Model Card for Model ID

A specialized language model powering LendWell’s Broker Assist platform, designed to help mortgage brokers query structured application data, retrieve relevant documents, run compliance checks, and trigger domain-specific actions.

Model Details

Model Description

This decoder-only LLM interprets JSON-formatted mortgage application data, performs semantic retrieval via RAG, and orchestrates business tools (e.g., document validation, compliance checks, email reminders) through LangChain agents and a custom MCPChain orchestrator.

Developed by: LendWell limited
Funded by [optional]: LendWell limited
Shared by [optional]: LendWell limited
Model type: Decoder-only autoregressive LLM
Language(s) (NLP): English
License: Apache-2.0
Finetuned from model [optional]: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T

Model Sources [optional]

Repository: [More Information Needed]
Paper [optional]: [More Information Needed]
Demo [optional]: Integrated in LendWell Broker Assist UI

Uses

Direct Use

Answer broker questions about application status (e.g., “Has Abhishek Khanna submitted his payslip?”)
Retrieve missing-document lists or timelines from JSON fields
Trigger business actions like sending follow-up emails or running compliance scripts

Downstream Use [optional]

[More Information Needed]

Out-of-Scope Use

Automated underwriting or credit-scoring decisions without human review

General open-domain Q&A outside of mortgage application context

Legal or compliance judgments without broker confirmation

Bias, Risks, and Limitations

The model reflects biases in the training corpus of internal mortgage applications. It depends on well-formed JSON inputs—ambiguous or misspelled queries can lead to incorrect retrieval. Overreliance on LLM outputs for compliance or legal decisions is unsafe.

Recommendations

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

How to Get Started with the Model

from transformers import pipeline

qa = pipeline( "text-generation", model="speed222/mortgage-tiny-llama-qlora" )

print(qa("Show me the status of the bank statement for application X123"))

Training Details

Training Data

Public: mortgage consumer complaints.

Training Procedure

Preprocessing [optional]

Flatten JSON into labeled text

Training Hyperparameters

Training regime: fp16 mixed precision, Epochs: 3 , Training steps: ~100,000 , Context window: 4,096 tokens

Speeds, Sizes, Times [optional]

[More Information Needed]

Evaluation

Testing Data, Factors & Metrics

Testing Data

[More Information Needed]

Factors

[More Information Needed]

Metrics

[More Information Needed]

Results

[More Information Needed]

Summary

Model Examination [optional]

[More Information Needed]

Technical Specifications [optional]

Model Architecture and Objective

[More Information Needed]

Compute Infrastructure

[More Information Needed]

Hardware

[More Information Needed]

Software

[More Information Needed]

Citation [optional]

BibTeX:

@misc{lendwell2025mortgage, title = {LendWell Mortgage Assistant LLM}, author = {LendWell}, year = {2025}, howpublished = {\url{https://huggingface.co/speed222/mortgage-tiny-llama-qlora/}}, note = {Version 1.0} }

APA: LendWell. (2025). LendWell Mortgage Assistant LLM [Computer software]. https://huggingface.co/speed222/mortgage-tiny-llama-qlora/

Glossary [optional]

[More Information Needed]

More Information [optional]

[More Information Needed]

Model Card Authors [optional]

Abhishek Khanna (Head of AI, LendWell)

Model Card Contact

Email: [email protected] Website: https://lendwell.ie