Project: LabLens Health Insights API One-line summary: A suite of FastAPI microservices that converts raw blood/lab biomarker reports into AI-generated clinical insights, follow-up questions, personalized recommendations, and test suggestions.
- Domain: Data Science (Digital Health / Applied LLMs in Healthcare)
- Role framing: Backend AI/ML engineer and LLM application developer. Legitimate claims: designed and built production-style FastAPI services for clinical biomarker analysis; engineered LLM prompting and orchestration with a multi-provider fallback strategy; implemented a statistical + machine-learning pattern-detection layer over lab data; deployed multi-service architecture on Render.
This is a digital-health / biological-age and longevity oriented analysis platform: it ingests comprehensive lab panels (100+ biomarkers) and produces functional-medicine style clinical reasoning, category summaries, and recommendations. Because the analysis is AI-driven, such a product would carry meaningful clinical-validation and regulatory considerations before real-world diagnostic use.
Python 3.9 · FastAPI · Uvicorn · Pydantic · OpenAI GPT-4o · Google Gemini (REST) · scikit-learn (RandomForest, IsolationForest, DBSCAN) · NumPy · python-dotenv · Render (multi-service deploy)
- Built a suite of four independently deployable FastAPI microservices that transform comprehensive lab reports (100+ biomarkers) into structured clinical notes, follow-up questions, personalized diet/activity/supplement recommendations, and follow-up test suggestions.
- Engineered a resilient LLM layer using OpenAI GPT-4o with automatic Google Gemini fallback, rate limiting, and robust parsing of model JSON output, ensuring graceful degradation when a provider fails.
- Designed a hybrid Dynamic Pattern Detector combining statistical biomarker correlations, clinically meaningful ratio analysis (Triglyceride/HDL, Cholesterol/HDL, BUN/Creatinine, Neutrophil/Lymphocyte), and an adaptive scikit-learn ML layer (RandomForest, IsolationForest, DBSCAN) with online confidence scoring to enrich LLM prompts.
- Implemented a two-step question-generation pipeline that first produces contextual follow-up questions from abnormal/borderline biomarkers, then generates patient-friendly multiple-choice answer options for each.
- Mapped biomarkers across nine clinical categories (longevity, metabolic, cardiovascular, hormonal, immune, nutritional, organ health, performance, preventive) to produce targeted, category-level health summaries and reduced manual lab-interpretation effort by
[add metric: e.g. ~X% faster review].
Python · FastAPI · REST API design · Microservices architecture · LLM application development · Prompt engineering · OpenAI API · Google Gemini API · Multi-provider LLM orchestration & fallback · Pydantic data modeling · scikit-learn · Anomaly detection · Clustering · Statistical analysis · Healthcare / clinical data · Biomarker analysis · JSON data pipelines · Cloud deployment (Render) · Environment/secrets management