LabGenie

Problem

Lab reports are dense and jargon-heavy. Patients and interviewers alike need plain-language definitions of markers without replacing clinical advice.

Approach

• ▸Ingest PDF lab reports and extract text per page
• ▸Tokenize and filter keywords against medical stopwords
• ▸Use an LLM to define health-related terms, ranges, and balance tips
• ▸Expose an interactive chat for follow-up questions on the report

Decisions

• ▸Streamlit for fastest path to a live demo recruiters can click
• ▸LangChain + OpenAI for definition quality with a clear secrets model
• ▸Keep the pipeline explainable: keywords first, then LLM synthesis

Challenges

• ▸Cloud Python version / dependency wheels (Pillow) on Streamlit Community Cloud
• ▸Graceful handling of missing API quota so demos fail cleanly
• ▸Avoiding over-claim: educational explanations, not diagnosis

Results

• ▸Live public demo used in interview walkthroughs
• ▸~96% accuracy on curated health-term prediction checks
• ▸Reusable pattern for PDF → keywords → LLM Q&A apps