Prototype HR chatbot using DoRA fine-tuning on synthetic Q&A data. System produces explainable, plain-language responses with source attribution for policy queries.
Evaluation via LLM-as-a-Judge using a Prometheus-based framework, achieving strong correlation with human ratings (r = 0.82): fine-tuning performance improved from 1.2 → 3.0 (1–5 scale).
project.ipynb— full end-to-end pipeline (data, training, evaluation, inference)report.pdf— final reportproposal.pdf— original proposalpresentation.pptx— overview slides