resume_corrected.json

{
  "CONTACT INFO": {
    "phone": "+1(984)325-5829",
    "location": "San Diego",
    "email": "[email protected]",
    "github_user": "JaynilJaiswal",
    "linkedin_user": "jaynil-jaiswal",
    "full_text_original": "+1(984)325-5829\nSan Diego\[email protected]\nJaynil Jaiswal GitHub: JaynilJaiswal\nLinkedIn: jaynil-jaiswal"
  },
  "EDUCATION": [
    {
      "institution": "University of California, San Diego",
      "dates": "Sep 2023 - Mar 2025",
      "degree": "Master of Science",
      "major": "Computer Science",
      "full_text_original": "University of California, San Diego Sep 2023 - Mar 2025\nMaster of Science: Computer Science"
    },
    {
      "institution": "Indian Institute of Technology, Roorkee",
      "dates": "Jul 2017 - Jun 2021",
      "degree": "Bachelor of Technology",
      "major": "Computer Science & Engineering",
      "full_text_original": "Indian Institute of Technology, Roorkee Jul 2017 - Jun 2021\nBachelor of Technology: Computer Science & Engineering"
    }
  ],
  "WORK EXPERIENCE": [
    {
      "title": "Graduate Student Researcher",
      "organization": "WiFire Lab, UCSD",
      "location": "San Diego",
      "dates": "April 2024 - Present",
      "description_points": [
        "Developed a deep learning pipeline to emulate the QUIC-Fire physics-based simulator, enabling faster modeling of fire spread under varying environmental conditions using a 234GB grassland dataset.",
        "Proposed a novel convex-hull-based omnidirectional Rate of Spread (RoS) calculation strategy and implemented a custom dataloader to train ConvLSTM without memory bottlenecks.",
        "Achieved a 66.67% reduction in MSE using a Physics-Guided loss in TensorFlow, outperforming baseline models."
      ],
      "full_text_original": "Graduate Student Researcher \u2014 WiFire Lab, UCSD April 2024 - Present\n\u2022 Developed a deep learning pipeline to emulate the QUIC-Fire physics-based simulator, enabling faster modeling\nof fire spread under varying environmental conditions using a 234GB grassland dataset.\n\u2022 Proposed a novel convex-hull-based omnidirectional Rate of Spread (RoS) calculation strategy and implemented\na custom dataloader to train ConvLSTM without memory bottlenecks.\n\u2022 Achieved a 66.67% reduction in MSE using a Physics-Guided loss in TensorFlow, outperforming baseline\nmodels.",
      "relevant_skills": [
        "ConvLSTM",
        "Data Pipelines",
        "Deep Learning",
        "ROS",
        "Simulation",
        "Tensorflow"
      ]
    },
    {
      "title": "Software Engineer IC2",
      "organization": "Oracle Corp.",
      "location": "Bangalore, India",
      "dates": "July 2021 - September 2023",
      "description_points": [
        "Worked on the Dynamic Routing Gateway (DRG) service to implement a route filtering feature allowing CIDR-based aggregation across attachments. Involved changes across API layers, Swagger spec, and SDKs (Terraform, Go, Python, CLI), along with relevant tests and monitoring setups.",
        "Deployed code changes globally across 18+ OCI regions using internal DevOps tooling with Docker and Terraform scripts.",
        "Built a Go-based tool named Fleet Visualizer and Restarter for real-time monitoring and selective restarts of congested DRG routers. Helped reduce incident response time and improved fleet health visibility across availability domains.",
        "Documented DRG connection path behavior (e.g., VCN peering, FastConnect, IPsec) by collaborating with senior engineers. Created internal Confluence documentation and runbooks to support on-call debugging and operational clarity.",
        "Developed an ETL pipeline for collecting and storing router metrics into Redis and Aerospike, capable of handling up to 1 million updates/sec.",
        "Actively maintained and enhanced the Terraform provider for DRG, adding new feature flags and ensuring backward compatibility. Managed canary deployments and telemetry alarms to track regressions."
      ],
      "full_text_original": "Software Engineer IC2 \u2014 Oracle Corp., Bangalore, India July 2021 - September 2023\n\u2022 Worked on the Dynamic Routing Gateway (DRG) service to implement a route filtering feature allowing\nCIDR-based aggregation across attachments. Involved changes across API layers, Swagger spec, and SDKs\n(Terraform, Go, Python, CLI), along with relevant tests and monitoring setups.\n\u2022 Deployed code changes globally across 18+ OCI regions using internal DevOps tooling with Docker and\nTerraform scripts.\n\u2022 Built a Go-based tool named Fleet Visualizer and Restarter for real-time monitoring and selective restarts\nof congested DRG routers. Helped reduce incident response time and improved fleet health visibility across\navailability domains.\n\u2022 Documented DRG connection path behavior (e.g., VCN peering, FastConnect, IPsec) by collaborating with\nsenior engineers. Created internal Confluence documentation and runbooks to support on-call debugging and\noperational clarity.\n\u2022 Developed an ETL pipeline for collecting and storing router metrics into Redis and Aerospike, capable of\nhandling up to 1 million updates/sec.\n\u2022 Actively maintained and enhanced the Terraform provider for DRG, adding new feature flags and ensuring\nbackward compatibility. Managed canary deployments and telemetry alarms to track regressions.",
      "relevant_skills": [
        "AeroSpike",
        "CIDR",
        "Confluence",
        "Data Pipelines",
        "DevOps",
        "Docker",
        "ETL Pipelines",
        "Go",
        "Networking",
        "Oracle Cloud Infrastructure",
        "Python",
        "RESTful APIs",
        "ROS",
        "Redis",
        "SDK Development",
        "Terraform"
      ]
    },
    {
      "title": "MLOps Co-op",
      "organization": "Proton AutoML (Acquired by Cliently)",
      "location": "Remote",
      "dates": "May 2020 - February 2021",
      "description_points": [
        "Architected and implemented robust cloud infrastructure on AWS for 160GB/server of data per EC2 instance.",
        "Engineered data pipelines that seamlessly integrated data from various sources such as Amazon RDS, Snowflake, Google Drive, and Dropbox.",
        "Designed and deployed scalable cloud infrastructure on AWS for AutoML application, implementing Kubernetes-based orchestration to support high-memory model workloads.",
        "Containerized the web application and configured networking with AWS load balancers for public access and reliability.",
        "Improved model pipeline performance by 1.3x using task parallelism with DaskML for distributed data processing."
      ],
      "full_text_original": "MLOps Co-op \u2014 Proton AutoML (Acquired by Cliently),Remote May 2020 - February 2021\n\u2022 Architected and implemented robust cloud infrastructure on AWS for 160GB/server of data per EC2 instance.\n\u2022 Engineered data pipelines that seamlessly integrated data from various sources such as Amazon RDS, Snowflake,\nGoogle Drive, and Dropbox.\n\u2022 Designed and deployed scalable cloud infrastructure on AWS for AutoML application, implementing Kubernetes-\nbased orchestration to support high-memory model workloads.\n\u2022 Containerized the web application and configured networking with AWS load balancers for public access and\nreliability.\n\u2022 Improved model pipeline performance by 1.3x using task parallelism with DaskML for distributed data pro-\ncessing.",
      "relevant_skills": [
        "AWS",
        "Cloud Computing",
        "Dask",
        "DaskML",
        "Data Pipelines",
        "Kubernetes",
        "Snowflake"
      ]
    },
    {
      "title": "Research Intern",
      "organization": "Video Analytics Lab, IISc Bangalore",
      "location": "Bangalore",
      "dates": "November 2019 - January 2020",
      "description_points": [
        "Experimented with the fusion techniques of LDR images using Autoencoder architecture to de-ghost HDR images capturing motion. Achieved PSNR of 41db using BGU upsampling of LDR representation of the final merged image.",
        "Employed NVidia DGX-1 GPU cluster to conduct model training on 40,960 cores."
      ],
      "full_text_original": "Research Intern \u2014 Video Analytics Lab, IISc Bangalore November 2019 - January 2020\n\u2022 Experimented with the fusion techniques of LDR images using Autoencoder architecture to de-ghost HDR\nimages capturing motion. Achieved PSNR of 41db using BGU upsampling of LDR representation of the final\nmerged image.\n\u2022 Employed NVidia DGX-1 GPU cluster to conduct model training on 40,960 cores.",
      "relevant_skills": [
        "Autoencoder",
        "GPU",
        "GPU Programming"
      ]
    },
    {
      "title": "Student Researcher",
      "organization": "Machine Vision & Intelligence Lab, IIT Roorkee",
      "location": "Roorkee",
      "dates": "January 2019 - June 2019",
      "description_points": [
        "Surveyed the existing state-of-the-art text-to-speech evaluation techniques, identified research gaps and developed an objective metric to automatically evaluate the human-ness of machine-synthesized speech using Conditional Generative Adversarial Networks.",
        "Developed an ensemble evaluation framework using Conditional GANs to automatically score speech quality in TTS systems, closely correlating with human vMOS scores from models like Tacotron2, DeepVoice3, and Wavenet.",
        "Published my research as A Generative Adversarial Network Based Ensemble Technique for Automatic Evaluation of Machine Synthesized Speech, ACPR 2019. Lecture Notes in Computer Science(), vol 12047. Springer"
      ],
      "full_text_original": "Student Researcher \u2014 Machine Vision & Intelligence Lab, IIT Roorkee January 2019 - June 2019\n\u2022 Surveyed the existing state-of-the-art text-to-speech evaluation techniques, identified research gaps and de-\nveloped an objective metric to automatically evaluate the human-ness of machine-synthesized speech using\nConditional Generative Adversarial Networks.\n\u2022 Developed an ensemble evaluation framework using Conditional GANs to automatically score speech quality\nin TTS systems, closely correlating with human vMOS scores from models like Tacotron2, DeepVoice3, and\nWavenet.\n\u2022 Published my research as A Generative Adversarial Network Based Ensemble Technique for Automatic Eval-\nuation of Machine Synthesized Speech, ACPR 2019. Lecture Notes in Computer Science(), vol 12047. Springer",
      "relevant_skills": [
        "Generative Adversarial Networks (GANs)",
        "Speech Processing",
        "Text-to-speech"
      ]
    }
  ],
  "PROJECTS": [
    {
      "name": "Sales Data Warehouse & BI Dashboard Pipeline",
      "technologies": [
        "PostgreSQL",
        "SQL (CTEs, Window Functions, GROUPING SETS)",
        "Python",
        "Looker Studio",
        "cron"
      ],
      "description_points": [
        "Designed and deployed a full-stack data warehouse and BI reporting solution to support executive decision-making using regional sales KPIs.",
        "Built a star schema with fact and dimension tables to consolidate daily transactional data from multiple operational sources into a centralized PostgreSQL data warehouse.",
        "Developed high-performance ETL pipelines using pure SQL to process over 10M rows/day, including data normalization, surrogate key generation, and incremental data loads.",
        "Applied advanced SQL techniques such as CTEs and window functions (e.g., ROW NUMBER, SUM OVER PARTITION) for deduplication, trend analysis, and time-based aggregation.",
        "Tuned database performance with composite indexing on (dateid, product id) and monthly partitioning of large fact tables, reducing average query latency from ~5s to <500ms.",
        "Created and scheduled materialized views for daily and monthly rollups to power a real-time interactive dashboard built in Looker Studio.",
        "Implemented multi-dimensional analysis using GROUPING SETS, ROLLUP, and CUBE to enable drilldowns by region, product category, and time hierarchy.",
        "Automated refreshes of materialized views using cron jobs and optimized query execution plans to maintain sub-second dashboard responsiveness for over 100 concurrent users."
      ],
      "full_text_original": "Sales Data Warehouse & BI Dashboard Pipeline PostgreSQL, SQL (CTEs, Window Functions, GROUPING\nSETS), Python, Looker Studio, cron\n\u2022 Designed and deployed a full-stack data warehouse and BI reporting solution to support executive decision-\nmaking using regional sales KPIs.\n\u2022 Built a star schema with fact and dimension tables to consolidate daily transactional data from multiple\noperational sources into a centralized PostgreSQL data warehouse.\n\u2022 Developed high-performance ETL pipelines using pure SQL to process over 10M rows/day, including data\nnormalization, surrogate key generation, and incremental data loads.\n\u2022 Applied advanced SQL techniques such as CTEs and window functions (e.g., ROW NUMBER, SUM OVER PARTITION)\nfor deduplication, trend analysis, and time-based aggregation.\n\u2022 Tuned database performance with composite indexing on (dateid, product id) and monthly partitioning\nof large fact tables, reducing average query latency from ~5s to <500ms.\n\u2022 Created and scheduled materialized views for daily and monthly rollups to power a real-time interactive\ndashboard built in Looker Studio.\n\u2022 Implemented multi-dimensional analysis using GROUPING SETS, ROLLUP, and CUBE to enable\ndrilldowns by region, product category, and time hierarchy.\n\u2022 Automated refreshes of materialized views using cron jobs and optimized query execution plans to maintain\nsub-second dashboard responsiveness for over 100 concurrent users.",
      "relevant_skills": [
        "CTEs",
        "Data Pipelines",
        "Data Warehousing",
        "ETL Pipelines",
        "Looker Studio",
        "Materialized Views",
        "Partitioning",
        "PostgreSQL",
        "Python",
        "RAG",
        "SQL",
        "SQL (CTEs, Window Functions, GROUPING SETS)",
        "Star Schema",
        "Window Functions",
        "cron"
      ]
    },
    {
      "name": "Instruction-Following Fine-Tuning of FLAN-T5-XL and Mistral-7B",
      "technologies": [
        "FLAN-T5-XL",
        "Mistral-7B-Instruct",
        "LoRA",
        "4-bit quantization",
        "GPU"
      ],
      "description_points": [
        "Fine-tuned FLAN-T5-XL and Mistral-7B-Instruct on the Bitext customer-support dataset using LoRA and 4-bit quantization on a 6GB GPU.",
        "Achieved a 28% improvement in ROUGE-L and notable gains in BLEU and GPTScore over base models.",
        "Reduced memory usage by 60% with parameter-efficient tuning while preserving generation quality."
      ],
      "full_text_original": "Instruction-Following Fine-Tuning of FLAN-T5-XL and Mistral-7B\n\u2022 Fine-tuned FLAN-T5-XL and Mistral-7B-Instruct on the Bitext customer-support dataset using LoRA\nand 4-bit quantization on a 6GB GPU.\n\u2022 Achieved a 28% improvement in ROUGE-L and notable gains in BLEU and GPTScore over base models.\n\u2022 Reduced memory usage by 60% with parameter-efficient tuning while preserving generation quality.",
      "relevant_skills": [
        "4-bit quantization",
        "FLAN-T5-XL",
        "GPU",
        "GPU Programming",
        "LoRA",
        "Mistral-7B-Instruct",
        "Model Fine-Tuning",
        "Model Quantization"
      ]
    },
    {
      "name": "Movie Recommendation Engine Using ElasticSearch Graph",
      "technologies": [
        "Elasticsearch",
        "Elasticsearch Graph plugin",
        "MovieLens dataset"
      ],
      "description_points": [
        "Developed a movie recommendation engine using the ElasticSearch Graph plugin to analyze a dataset of 27,300 movies. The project processed over 138,000 user entities and handled more than 20 million ratings to generate personalized movie suggestions based on user preferences.",
        "Developed a recommendation engine using Elasticsearch Graph to analyze movie ratings and suggest personalized movie recommendations based on user preferences.",
        "Processed and indexed large-scale datasets from MovieLens into Elasticsearch, structuring data to identify patterns and connections among user preferences.",
        "Leveraged Elasticsearch Graph plugin to explore and visualize relationships between movies, optimizing recommendations using 'wisdom of crowds' properties.",
        "Addressed data biases and enhanced recommendation diversity through advanced graph analysis and customized diversification techniques."
      ],
      "full_text_original": "Movie Recommendation Engine Using ElasticSearch Graph\n\u2022 Developed a movie recommendation engine using the ElasticSearch Graph plugin to analyze a dataset of 27,300\nmovies. The project processed over 138,000 user entities and handled more than 20 million ratings to generate\npersonalized movie suggestions based on user preferences.\n\u2022 Developed a recommendation engine using Elasticsearch Graph to analyze movie ratings and suggest personal-\nized movie recommendations based on user preferences.\n\u2022 Processed and indexed large-scale datasets from MovieLens into Elasticsearch, structuring data to identify\npatterns and connections among user preferences.\n\u2022 Leveraged Elasticsearch Graph plugin to explore and visualize relationships between movies, optimizing recom-\nmendations using \u201dwisdom of crowds\u201d properties.\n\u2022 Addressed data biases and enhanced recommendation diversity through advanced graph analysis and customized\ndiversification techniques.",
      "relevant_skills": [
        "Elasticsearch",
        "Elasticsearch Graph plugin",
        "MovieLens dataset",
        "RAG",
        "Recommender Systems"
      ]
    },
    {
      "name": "Matrix Multiplication on GeneSys Hardware Accelerator",
      "technologies": [
        "GeneSys hardware accelerator",
        "Systolic array",
        "Processing Elements (PEs)",
        "Custom ISA",
        "Compiler",
        "Iterator table",
        "Instruction scheduling",
        "Tiling"
      ],
      "description_points": [
        "Programmed the GeneSys hardware accelerator, utilizing a 4x4 systolic array of Processing Elements (PEs) to perform matrix multiplication on 3x3 and 7x7 matrices with optimized efficiency.",
        "Integrated GeneSys\u2019s custom Instruction Set Architecture (ISA) and compiler, which converts high-level neural network computations into a sequence of hardware-specific instructions.",
        "Leveraged the iterator table in the buffer to manage instruction positions in the computation loop, ensuring synchronized data flow across PEs during matrix operations.",
        "Employed instruction scheduling and tiling techniques to maximize the utilization of PEs for larger matrices, achieving high performance in neural network workload simulations."
      ],
      "full_text_original": "Matrix Multiplication on GeneSys Hardware Accelerator\n\u2022 Programmed the GeneSys hardware accelerator, utilizing a 4x4 systolic array of Processing Elements (PEs)\nto perform matrix multiplication on 3x3 and 7x7 matrices with optimized efficiency.\n\u2022 Integrated GeneSys\u2019s custom Instruction Set Architecture (ISA) and compiler, which converts high-level neural\nnetwork computations into a sequence of hardware-specific instructions.\n\u2022 Leveraged the iterator table in the buffer to manage instruction positions in the computation loop, ensuring\nsynchronized data flow across PEs during matrix operations.\n\u2022 Employed instruction scheduling and tiling techniques to maximize the utilization of PEs for larger matrices,\nachieving high performance in neural network workload simulations.",
      "relevant_skills": [
        "Compiler",
        "Custom ISA",
        "GeneSys hardware accelerator",
        "Instruction scheduling",
        "Iterator table",
        "Processing Elements (PEs)",
        "RAG",
        "ROS",
        "Systolic array",
        "Tiling"
      ]
    },
    {
      "name": "SLAM System with AprilTags, Kalman Filter, and ROS2",
      "technologies": [
        "ROS2",
        "AprilTags",
        "Kalman Filter",
        "tf (ROS2 transform library)"
      ],
      "description_points": [
        "Developed a ROS2-based SLAM system integrating AprilTag-based localization and a custom Kalman filter for robust landmark tracking.",
        "Utilized ROS2 tf to transform detected AprilTag poses from the camera frame to world coordinates, with the map origin fixed at (0,0).",
        "Implemented an efficient landmark management system using a single array indexed by AprilTag IDs to store Kalman filter objects.",
        "Enabled dynamic tracking by using the update step of the Kalman filter for detected landmarks and the prediction step to maintain accuracy during intermittent detections.",
        "Achieved a memory-efficient design by eliminating redundant data structures, ensuring real-time performance in dynamic environments."
      ],
      "full_text_original": "SLAM System with AprilTags, Kalman Filter, and ROS2\n\u2022 Developed a ROS2-based SLAM system integrating AprilTag-based localization and a custom Kalman\nfilter for robust landmark tracking.\n\u2022 Utilized ROS2 tf to transform detected AprilTag poses from the camera frame to world coordinates, with the\nmap origin fixed at (0,0).\n\u2022 Implemented an efficient landmark management system using a single array indexed by AprilTag IDs to\nstore Kalman filter objects.\n\u2022 Enabled dynamic tracking by using the update step of the Kalman filter for detected landmarks and the\nprediction step to maintain accuracy during intermittent detections.\n\u2022 Achieved a memory-efficient design by eliminating redundant data structures, ensuring real-time performance\nin dynamic environments.",
      "relevant_skills": [
        "AprilTags",
        "Kalman Filter",
        "ROS",
        "ROS2",
        "SLAM",
        "tf (ROS2 transform library)"
      ]
    },
    {
      "name": "Streaming ETL Pipeline using Kafka and Airflow",
      "technologies": [
        "Apache Kafka",
        "MySQL",
        "Apache Airflow",
        "Multiprocessing"
      ],
      "description_points": [
        "Automated a scalable, real-time ETL pipeline using Apache Kafka, MySQL, and Apache Airflow to ingest and process vehicle data for toll traffic analysis and real-time monitoring.",
        "Leveraged multiprocessing to efficiently handle over 1 million records, significantly enhancing data ingestion speed and optimizing workload distribution.",
        "Automated a scalable real-time ETL pipeline using Apache Kafka, MySQL, and Apache Airflow, orchestrating end-to-end data processing for toll traffic analysis.",
        "Leveraged multiprocessing to handle over 1 million records, significantly improving data ingestion speed and optimizing workload distribution across multiple consumers.",
        "Streamed vehicle data to Kafka topics, ensuring efficient data ingestion and storage in a MySQL database for real-time traffic monitoring and analysis."
      ],
      "full_text_original": "Streaming ETL Pipeline using Kafka and Airflow\n\u2022 Automated a scalable, real-time ETL pipeline using Apache Kafka, MySQL, and Apache Airflow to ingest and\nprocess vehicle data for toll traffic analysis and real-time monitoring.\n\u2022 Leveraged multiprocessing to efficiently handle over 1 million records, significantly enhancing data ingestion\nspeed and optimizing workload distribution.\n\u2022 Automated a scalable real-time ETL pipeline using Apache Kafka, MySQL, and Apache Airflow, or-\nchestrating end-to-end data processing for toll traffic analysis.\n\u2022 Leveraged multiprocessing to handle over 1 million records, significantly improving data ingestion speed\nand optimizing workload distribution across multiple consumers.\n\u2022 Streamed vehicle data to Kafka topics, ensuring efficient data ingestion and storage in a MySQL database for\nreal-time traffic monitoring and analysis.",
      "relevant_skills": [
        "Apache Airflow",
        "Apache Kafka",
        "Data Pipelines",
        "ETL Pipelines",
        "Kafka",
        "Multiprocessing",
        "MySQL",
        "RAG",
        "ROS",
        "SQL"
      ]
    },
    {
      "name": "RAG-based Code generation using LLM",
      "technologies": [
        "RAG",
        "OLLama",
        "LangChain",
        "ChromaDB",
        "CMU CoNaLa dataset",
        "Python"
      ],
      "description_points": [
        "Developed a Retrieval Augmented Generation (RAG) system for code generation by leveraging OLLama for LLM integration and LangChain to orchestrate the retrieval and generation workflow.",
        "Utilized ChromaDB vector database to embed and persistently store over 590,000 JSON records, ensuring efficient and scalable data retrieval.",
        "Integrated the CMU CoNaLa dataset, comprising language and code pairs mined from StackOverflow, to generate accurate Python code snippets in response to natural language queries."
      ],
      "full_text_original": "RAG-based Code generation using LLM\n\u2022 Developed a Retrieval Augmented Generation (RAG) system for code generation by leveraging OLLama for\nLLM integration and LangChain to orchestrate the retrieval and generation workflow.\n\u2022 Utilized ChromaDB vector database to embed and persistently store over 590,000 JSON records, ensuring\nefficient and scalable data retrieval.\n\u2022 Integrated the CMU CoNaLa dataset, comprising language and code pairs mined from StackOverflow, to\ngenerate accurate Python code snippets in response to natural language queries.",
      "relevant_skills": [
        "CMU CoNaLa dataset",
        "ChromaDB",
        "LangChain",
        "OLLama",
        "Python",
        "RAG",
        "Vector Database"
      ]
    },
    {
      "name": "Optimized Matrix Multiplication on CUDA",
      "technologies": [
        "CUDA",
        "Nsight",
        "Shared Memory",
        "2D Tiling",
        "Memory Coalescing"
      ],
      "description_points": [
        "Implemented an efficient matrix multiplication algorithm using CUDA, leveraging shared memory for performance optimization.",
        "Developed a 2D tiling approach to maximize data reuse and minimize global memory access latency, enhancing computational throughput.",
        "Achieved significant performance gains, demonstrating high arithmetic intensity and efficient utilization of GPU resources.",
        "Validated the implementation using Nsight for performance profiling, achieving a throughput of 3386 GFLOPS with minimal DRAM bandwidth usage.",
        "Demonstrated proficiency in CUDA programming, memory coalescing, and performance optimization techniques for high-performance computing applications."
      ],
      "full_text_original": "Optimized Matrix Multiplication on CUDA\n\u2022 Implemented an efficient matrix multiplication algorithm using CUDA, leveraging shared memory for perfor-\nmance optimization.\n\u2022 Developed a 2D tiling approach to maximize data reuse and minimize global memory access latency, enhancing\ncomputational throughput.\n\u2022 Achieved significant performance gains, demonstrating high arithmetic intensity and efficient utilization of GPU\nresources.\n\u2022 Validated the implementation using Nsight for performance profiling, achieving a throughput of 3386 GFLOPS\nwith minimal DRAM bandwidth usage.\n\u2022 Demonstrated proficiency in CUDA programming, memory coalescing, and performance optimization tech-\nniques for high-performance computing applications.",
      "relevant_skills": [
        "2D Tiling",
        "CUDA",
        "GPU",
        "GPU Programming",
        "Memory Coalescing",
        "Nsight",
        "RAG",
        "Shared Memory",
        "Tiling"
      ]
    },
    {
      "name": "A Simple Approach to Black-box Adversarial Attacks",
      "organization": "University of California, San Diego",
      "technologies": [
        "MNIST",
        "CIFAR-10",
        "Autoencoder",
        "Transformer Encoder"
      ],
      "description_points": [
        "Developed a novel adversarial model that uses a specialized loss function combining reconstruction similarity and adversarial impact to generate adversarial examples that maintain their label but mislead the neural network.",
        "Evaluated the model on MNIST and CIFAR-10 datasets, demonstrating significant impairment of classifier accuracy, with a drop in accuracy from 99% to 23% on MNIST and from 90% to 41% on CIFAR-10.",
        "Implemented a 3-layer autoencoder architecture for the generator and a pretrained 3-layer digit classifier for MNIST.",
        "Addressed the balance between generating effective adversarial examples and preserving their original label by fine-tuning the hyperparameter \u03bb through empirical testing. Conducted experiments on textual data using a transformer encoder."
      ],
      "full_text_original": "A Simple Approach to Black-box Adversarial Attacks | University of California, San Diego\n\u2022 Developed a novel adversarial model that uses a specialized loss function combining reconstruction similarity and\nadversarial impact to generate adversarial examples that maintain their label but mislead the neural network.\n\u2022 Evaluated the model on MNIST and CIFAR-10 datasets, demonstrating significant impairment of classifier\naccuracy, with a drop in accuracy from 99% to 23% on MNIST and from 90% to 41% on CIFAR-10.\n\u2022 Implemented a 3-layer autoencoder architecture for the generator and a pretrained 3-layer digit classifier for\nMNIST.\n\u2022 Addressed the balance between generating effective adversarial examples and preserving their original label by fine-\ntuning the hyperparameter \u03bbthroughempiricaltesting.Conductedexperimentsontextualdatausingatransf ormerencodera",
      "relevant_skills": [
        "Adversarial Attacks",
        "Autoencoder",
        "CIFAR-10",
        "MNIST",
        "Model Fine-Tuning",
        "Transformer Encoder"
      ]
    },
    {
      "name": "Fleet Health Visualizer & Restarter",
      "organization": "Oracle India Pvt. Ltd.",
      "technologies": [
        "Go"
      ],
      "description_points": [
        "We made a ops tool to monitor availability of the service by various health metrics at a glance for different sources of traffic.",
        "We can also visualize traffic distribution over different fleets to see if there is congestion on a host.",
        "We can promptly respond to any incident and restore the availability by doing bulk restarts of the hosts."
      ],
      "full_text_original": "Fleet Health Visualizer & Restarter | Oracle India Pvt. Ltd.\n\u2022\u2022 We made a ops tool to monitor availability of the service by various health metrics at a glance for different\nsources of traffic.\n\u2022 We can also visualize traffic distribution over different fleets to see if there is congestion on a host.\n\u2022 We can promptly respond to any incident and restore the availability by doing bulk restarts of the hosts.",
      "relevant_skills": [
        "Go"
      ]
    },
    {
      "name": "Route Filtering",
      "organization": "Oracle India Pvt. Ltd.",
      "technologies": [
        "CIDR"
      ],
      "description_points": [
        "Created a flag to enable aggregating of dynamic routes rules using CIDR for virtual cloud network.",
        "This reduces the amount of route rules needed to store in our route table and enhance performance."
      ],
      "full_text_original": "Route Filtering | Oracle India Pvt. Ltd.\n\u2022 Created a flag to enable aggregating of dynamic routes rules using CIDR for virtual cloud network.\n\u2022 This reduces the amount of route rules needed to store in our route table and enhance performance.",
      "relevant_skills": [
        "CIDR"
      ]
    },
    {
      "name": "OLIVIA (Optimized Lightweight Intelligent Voice Interactive Assistant)",
      "technologies": [
        "Speech-to-text",
        "Text-to-speech",
        "NLU",
        "Chatbot",
        "Q&A",
        "Deep Punctuation",
        "Voice Authentication",
        "Text Summarization"
      ],
      "description_points": [
        "Developed a versatile voice assistant that can be tailored to different industries and carry out multiple tasks simultaneously.",
        "Implemented primary modules such as Speech-to-text, Text-to-speech, Natural-Language-Understanding (NLU) shell, Chatbot, Question And Answering, and Deep Punctuation.",
        "Integrated secondary modules like Snap/Clap activation, AdultFilter, Voice Authentication, and TextSummarization to enhance the assistant\u2019s performance and user experience.",
        "The feature pool included functionalities such as Time, Weather, Schedule List/Timetable, Play Music, Find Information, Send Message, Send Email, Play best videos related to a query, Handle Call and extract information or send a relevant information, and Trending news."
      ],
      "full_text_original": "OLIVIA (Optimized Lightweight Intelligent Voice Interactive Assistant)\n\u2022 Developed a versatile voice assistant that can be tailored to different industries and carry out multiple tasks\nsimultaneously.\n\u2022 Implemented primary modules such as Speech-to-text, Text-to-speech, Natural-Language-Understanding (NLU)\nshell, Chatbot, Question And Answering, and Deep Punctuation.\n\u2022 Integrated secondary modules like Snap/Clap activation, AdultFilter, Voice Authentication, and TextSumma-\nrization to enhance the assistant\u2019s performance and user experience.\n\u2022 The feature pool included functionalities such as Time, Weather, Schedule List/Timetable, Play Music, Find\nInformation, Send Message, Send Email, Play best videos related to a query, Handle Call and extract information\nor send a relevant information, and Trending news.",
      "relevant_skills": [
        "Chatbot",
        "Deep Punctuation",
        "NLU",
        "Q&A",
        "Speech Processing",
        "Speech-to-text",
        "Text Summarization",
        "Text-to-speech",
        "Voice Authentication"
      ]
    },
    {
      "name": "Few-Shot Unsupervised Image-to-Image Translation",
      "technologies": [
        "Generative Adversarial Networks (GANs)",
        "Attention mechanisms",
        "InceptionNet V3"
      ],
      "description_points": [
        "Applied Generative Adversarial Networks (GANs) for unsupervised image translation, innovatively extending the state-of-the-art model to enhance generalization across classes.",
        "Employed attention mechanisms, leveraging multiple sample images from various classes, to proficiently learn and adapt to new classes.",
        "Validated performance through InceptionNet V3 scores pretrained on the target class, achieving a notable 0.73 accuracy in Top-5 test solely with Vanilla-GAN."
      ],
      "full_text_original": "Few-Shot Unsupervised Image-to-Image Translation\n\u2022 Applied Generative Adversarial Networks (GANs) for unsupervised image translation, innovatively extending\nthe state-of-the-art model to enhance generalization across classes.\n\u2022 Employed attention mechanisms, leveraging multiple sample images from various classes, to proficiently learn\nand adapt to new classes.\n\u2022 Validated performance through InceptionNet V3 scores pretrained on the target class, achieving a notable\n0.73 accuracy in Top-5 test solely with Vanilla-GAN.",
      "relevant_skills": [
        "Attention mechanisms",
        "Generative Adversarial Networks (GANs)",
        "InceptionNet V3",
        "RAG",
        "ROS"
      ]
    },
    {
      "name": "Cross Domain Sentiment Analysis",
      "organization": "IIT Roorkee",
      "technologies": [
        "Neural Networks",
        "Gradient Reversal Layer"
      ],
      "description_points": [
        "The approach implements this idea in the context of neural network architectures that are trained on labeled data from the source domain and unlabeled data from the target domain (no labeled target-domain data is necessary).",
        "We show that this adaptation behavior can be achieved in almost any feed-forward model by augmenting it with a few standard layers and a new gradient reversal layer."
      ],
      "full_text_original": "Cross Domain Sentiment Analysis \u2014 IIT Roorkee\n\u2022 The approach implements this idea in the context of neural network architectures that are trained on labeled\ndata from the source domain and unlabeled data from the target domain (no labeled target-domain data is\nnecessary).\n\u2022 We show that this adaptation behavior can be achieved in almost any feed-forward model by augmenting it\nwith a few standard layers and a new gradient reversal layer.",
      "relevant_skills": [
        "Gradient Reversal Layer",
        "Neural Networks",
        "ROS"
      ]
    },
    {
      "name": "Data augmenting using Variational Auto-Encoder",
      "technologies": [
        "Variational Auto-Encoder (VAE)",
        "Tensorflow",
        "MNIST dataset"
      ],
      "description_points": [
        "Implementated Variational Auto-Encoder using Tensorflow framework. Used MNIST dataset to train the network to learn representations of different numbers in many handwritings.",
        "I used the trained encoder to interpolate features of existing images to decode new images."
      ],
      "full_text_original": "Data augmenting using Variational Auto-Encoder\n\u2022 Implementated Variational Auto-Encoder using Tensorflow framework. Used MNIST dataset to train the net-\nwork to learn representations of different numbers in many handwritings.\n\u2022 I used the trained encoder to interpolate features of existing images to decode new images.",
      "relevant_skills": [
        "MNIST",
        "MNIST dataset",
        "Tensorflow",
        "Variational Auto-Encoder (VAE)"
      ]
    },
    {
      "name": "AEREM",
      "organization": "Microsoft Code.Fun.Do 2017",
      "technologies": [
        "C++",
        "OpenCV",
        "OCR"
      ],
      "description_points": [
        "An application written in C++ which make use of OpenCV to track motion of a pointer in front a camera and recognize the character written in air in front of it using OCR. The prototype contained capability to record letters written in air in front of a camera."
      ],
      "full_text_original": "AEREM \u2014 Microsoft Code.Fun.Do 2017\n\u2022 An application written in C++ which make use of OpenCV to track motion of a pointer in front a camera and\nrecognize the character written in air in front of it using OCR. The prototype contained capability to record\nletters written in air in front of a camera.",
      "relevant_skills": [
        "C",
        "C++",
        "OCR",
        "OpenCV",
        "ROS"
      ]
    }
  ],
  "ONLINE CERTIFICATIONS": [
    {
      "name": "IBM Data Engineering Professional Certificate",
      "issuer": "Coursera",
      "date": "Apr 2025",
      "description_points": [
        "Covered relational databases, SQL, data warehouses, ETL pipelines, and NoSQL systems.",
        "Built and managed data pipelines using tools like Apache Airflow, Cloud Object Storage, and Python.",
        "Focused on scalable data architecture, data wrangling, and operationalizing data engineering workflows."
      ],
      "full_text_original": "IBM Data Engineering Professional Certificate \u2013 Coursera Apr 2025\n\u2022 Covered relational databases, SQL, data warehouses, ETL pipelines, and NoSQL systems.\n\u2022 Built and managed data pipelines using tools like Apache Airflow, Cloud Object Storage, and Python.\n\u2022 Focused on scalable data architecture, data wrangling, and operationalizing data engineering workflows.",
      "relevant_skills": [
        "Apache Airflow",
        "Data Pipelines",
        "Data Warehousing",
        "ETL Pipelines",
        "NoSQL Databases",
        "Python",
        "RAG",
        "SQL"
      ]
    },
    {
      "name": "Mathematics for Machine Learning",
      "issuer": "Coursera",
      "date": "May 2020",
      "description_points": [
        "Covered Linear Algebra, Multivariate Calculus, and Principal Component Analysis.",
        "Focused on mathematical foundations for data science and machine learning."
      ],
      "full_text_original": "Mathematics for Machine Learning \u2013 Coursera May 2020\n\u2022 Covered Linear Algebra, Multivariate Calculus, and Principal Component Analysis.\n\u2022 Focused on mathematical foundations for data science and machine learning.",
      "relevant_skills": []
    },
    {
      "name": "Mathematics for Machine Learning",
      "issuer": "Coursera",
      "date": "May 2020",
      "description_points": [
        "Covered Linear Algebra, Multivariate Calculus, and Principal Component Analysis.",
        "Focused on mathematical foundations for data science and machine learning."
      ],
      "full_text_original": "Mathematics for Machine Learning \u2013 Coursera May 2020\n\u2022 Covered Linear Algebra, Multivariate Calculus, and Principal Component Analysis.\n\u2022 Focused on mathematical foundations for data science and machine learning.",
      "relevant_skills": []
    },
    {
      "name": "Deep Learning Specialization",
      "issuer": "Coursera",
      "date": "June 2019",
      "description_points": [
        "Courses completed:",
        "\u2013 Neural Networks and Deep Learning",
        "\u2013 Improving Deep Neural Networks: Hyperparameter Tuning, Regularization and Optimization",
        "\u2013 Structuring Machine Learning Projects",
        "\u2013 Convolutional Neural Networks",
        "Gained practical knowledge in deep learning and neural network architectures."
      ],
      "full_text_original": "Deep Learning Specialization \u2013 Coursera June 2019\n\u2022 Courses completed:\n\u2013 Neural Networks and Deep Learning\n\u2013 Improving Deep Neural Networks: Hyperparameter Tuning, Regularization and Optimization\n\u2013 Structuring Machine Learning Projects\n\u2013 Convolutional Neural Networks\n\u2022 Gained practical knowledge in deep learning and neural network architectures.",
      "relevant_skills": [
        "Deep Learning",
        "Neural Networks"
      ]
    },
    {
      "name": "Getting Started with SAS Programming",
      "issuer": "Coursera",
      "date": "May 2020",
      "description_points": [
        "Learned foundational skills in SAS programming for data analysis."
      ],
      "full_text_original": "Getting Started with SAS Programming \u2013 Coursera May 2020\n\u2022 Learned foundational skills in SAS programming for data analysis.",
      "relevant_skills": []
    },
    {
      "name": "Python for Data Science, AI Development",
      "issuer": "Coursera",
      "date": "October 2019",
      "description_points": [
        "Acquired proficiency in Python programming for data science and AI applications."
      ],
      "full_text_original": "Python for Data Science, AI Development \u2013 Coursera October 2019\n\u2022 Acquired proficiency in Python programming for data science and AI applications.",
      "relevant_skills": [
        "Python"
      ]
    }
  ],
  "COURSES": [
    {
      "code": "CSE 256",
      "name": "Statistical Natural Language Processing",
      "institution": "UC San Diego",
      "term": "Spring 2024",
      "description_points": [
        "Introduction to Natural Language Processing (NLP) focusing on deep learning methods and language model pre-training.",
        "Covered techniques include classification, feedforward neural networks, attention mechanisms, pre-trained language models (BERT, GPT), and structured models.",
        "Focus on NLP tasks such as semantics, question answering, and applications like machine translation."
      ],
      "full_text_original": "CSE 256: Statistical Natural Language Processing \u2013 UC San Diego Spring 2024\n\u2022 Introduction to Natural Language Processing (NLP) focusing on deep learning methods and language model\npre-training.\n\u2022 Covered techniques include classification, feedforward neural networks, attention mechanisms, pre-trained lan-\nguage models (BERT, GPT), and structured models.\n\u2022 Focus on NLP tasks such as semantics, question answering, and applications like machine translation.",
      "relevant_skills": [
        "Attention mechanisms",
        "Deep Learning",
        "Natural Language Processing (NLP)",
        "Neural Networks"
      ]
    },
    {
      "code": "CSE 260",
      "name": "Parallel Computation",
      "institution": "UC San Diego",
      "term": "Spring 2024",
      "description_points": [
        "Exploration of parallel computation and the challenges of exploiting parallelism in modern computer systems.",
        "Topics include performance estimation, parallel architecture, vectorization, GPUs, and large-scale supercomputers."
      ],
      "full_text_original": "CSE 260: Parallel Computation \u2013 UC San Diego Spring 2024\n\u2022 Exploration of parallel computation and the challenges of exploiting parallelism in modern computer systems.\n\u2022 Topics include performance estimation, parallel architecture, vectorization, GPUs, and large-scale supercom-\nputers.",
      "relevant_skills": [
        "GPU Programming",
        "LoRA"
      ]
    },
    {
      "code": "CSE 234",
      "name": "Data Systems for ML",
      "institution": "UC San Diego",
      "term": "Winter 2024",
      "description_points": [
        "In-depth analysis of Classical ML Training at Scale, covering seminal papers such as Parameter Server, XGBoost, MADLib, MLlib, Spark ML, Mahout, Graphlab, AWS Sagemaker.",
        "Exploration of cutting-edge DL Systems, including Distributed PyTorch, Alpa, TVM, Tensorflow, FSDP, Horovod, Pathways, Monolith.",
        "Thorough examination of ML deployment intricacies, MLOps frameworks, and emerging concepts like VLLM, Federated ML, LLMOps, LangChain, LlamaIndex."
      ],
      "full_text_original": "CSE 234: Data Systems for ML \u2013 UC San Diego Winter 2024\n\u2022 In-depth analysis of Classical ML Training at Scale, covering seminal papers such as Parameter Server, XGBoost,\nMADLib, MLlib, Spark ML, Mahout, Graphlab, AWS Sagemaker.\n\u2022 Exploration of cutting-edge DL Systems, including Distributed PyTorch, Alpa, TVM, Tensorflow, FSDP,\nHorovod, Pathways, Monolith.\n\u2022 Thorough examination of ML deployment intricacies, MLOps frameworks, and emerging concepts like VLLM,\nFederated ML, LLMOps, LangChain, LlamaIndex.",
      "relevant_skills": [
        "AWS",
        "LangChain",
        "LoRA",
        "PyTorch",
        "Spark",
        "Tensorflow"
      ]
    },
    {
      "code": "CSE 251B",
      "name": "Deep Learning",
      "institution": "UC San Diego",
      "term": "Winter 2024",
      "description_points": [
        "Developed a strong foundation in Neural Networks, spanning from basic concepts such as Multi-layer Perceptrons (MLP) and logistic regression to advanced topics like Transformers, Recurrent Neural-networks (RNNs), and Reinforcement Learning (RL).",
        "Implemented practical applications of neural networks, including single-layer networks, Convolutional Neural-networks (CNNs), Attention networks, and Generative Adversarial Networks (GANs)."
      ],
      "full_text_original": "CSE 251B: Deep Learning \u2013 UC San Diego Winter 2024\n\u2022 Developed a strong foundation in Neural Networks, spanning from basic concepts such as Multi-layer Perceptrons\n(MLP) and logistic regression to advanced topics like Transformers, Recurrent Neural-networks (RNNs), and\nReinforcement Learning (RL).\n\u2022 Implemented practical applications of neural networks, including single-layer networks, Convolutional Neural-\nnetworks (CNNs), Attention networks, and Generative Adversarial Networks (GANs).",
      "relevant_skills": [
        "Deep Learning",
        "Generative Adversarial Networks (GANs)",
        "Neural Networks"
      ]
    },
    {
      "code": "CSE 202",
      "name": "Design and Analysis of Algorithms",
      "institution": "UC San Diego",
      "term": "Winter 2024",
      "description_points": [
        "Executed formalization of computational problems through techniques such as reductions between problems, graph search, and greedy algorithms.",
        "Took charge of a course project centered on formalizing computational problems in solving games like Scrabble and innovating algorithms to address them."
      ],
      "full_text_original": "CSE 202: Design and Analysis of Algorithms \u2013 UC San Diego Winter 2024\n\u2022 Executed formalization of computational problems through techniques such as reductions between problems,\ngraph search, and greedy algorithms.\n\u2022 Took charge of a course project centered on formalizing computational problems in solving games like Scrabble\nand innovating algorithms to address them.",
      "relevant_skills": []
    },
    {
      "code": "CSE 250A",
      "name": "Probabilistic Reasoning & Learning",
      "institution": "UC San Diego",
      "term": "Fall 2023",
      "description_points": [
        "Implemented probabilistic methods for reasoning and decision-making under uncertainty, focusing on inference and learning in directed probabilistic graphical models.",
        "Applied expertise to enhance prediction and planning in Markov decision processes, contributing to applications in computer vision, robotics, speech recognition, natural language processing, and information retrieval."
      ],
      "full_text_original": "CSE 250A: Probabilistic Reasoning & Learning \u2013 UC San Diego Fall 2023\n\u2022 Implemented probabilistic methods for reasoning and decision-making under uncertainty, focusing on inference\nand learning in directed probabilistic graphical models.\n\u2022 Applied expertise to enhance prediction and planning in Markov decision processes, contributing to applications\nin computer vision, robotics, speech recognition, natural language processing, and information retrieval.",
      "relevant_skills": [
        "Computer Vision",
        "Natural Language Processing (NLP)",
        "Speech Processing"
      ]
    },
    {
      "code": "CSE 258",
      "name": "Recommender Systems & Web Mining",
      "institution": "UC San Diego",
      "term": "Fall 2023",
      "description_points": [
        "Mastered contemporary methods in recommender systems, data mining, and predictive analytics during an intensive graduate course.",
        "Proficient in handling temporal changes and sequential data, with specialized knowledge in visual recommendation systems, particularly in the context of fashion item recommendation."
      ],
      "full_text_original": "CSE 258: Recommender Systems & Web Mining \u2013 UC San Diego Fall 2023\n\u2022 Mastered contemporary methods in recommender systems, data mining, and predictive analytics during an\nintensive graduate course.\n\u2022 Proficient in handling temporal changes and sequential data, with specialized knowledge in visual recommen-\ndation systems, particularly in the context of fashion item recommendation.",
      "relevant_skills": [
        "Data Mining"
      ]
    }
  ],
  "EXTRACURRICULAR ACTIVITIES": [
    {
      "organization": "Vision & Language Group \u2014 ACM Chapter, IIT Roorkee",
      "description_points": [
        "Participated in group discussions and paper presentations.",
        "Used to hold small lectures on basic topics of Deep Learning."
      ],
      "full_text_original": "Vision & Language Group \u2014 ACM Chapter, IIT Roorkee\nParticipated in group discussions and paper presentations. Used to hold small lectures on basic topics of Deep\nLearning."
    }
  ]
}