ClinSumm: LoRA-Tuned Clinical Note Summarizer
A parameter-efficient fine-tuned language model that summarizes de-identified clinical notes into structured, clinician-readable briefs.
Overview
ClinSumm applies the LoRA fine-tuning technique from my reproduction work to a 1.3B-parameter open language model, adapting it to summarize long de-identified clinical notes into a structured problem list, medications, and follow-up plan, viewable through a lightweight Streamlit interface.
Architecture
LoRA adapters (rank 8) on the query/value projections of a frozen base decoder model; a retrieval step surfaces relevant prior note sections before summarization to keep context length manageable.
Tech Stack
Screenshots
Lessons Learned
Prompt structure mattered more than expected — explicit section headers in the target format cut hallucinated medication entries substantially compared to free-form summarization prompts. Evaluation required a clinician-style rubric since ROUGE scores correlated poorly with perceived summary quality.
ls -la ./related-projects/
Edge-NIDS: On-Device Intrusion Detection for IoT
A quantized CNN intrusion-detection model deployed to an ARM Cortex-M microcontroller for real-time IoT traffic monitoring without cloud connectivity.
FedDiagnose: Cross-Hospital Federated Diagnostic Modeling
A simulated multi-hospital federated learning platform for chest X-ray classification, keeping patient imaging data local to each institution.
ReproBench: Automated Paper Reproduction Scorecard
A tooling suite that runs reproduced model checkpoints against held-out benchmarks and auto-generates a comparison scorecard against reported paper results.