NeuroLens AI

Inspiration

People don't experience health changes as neat data points — they notice them as scattered moments: a foggy afternoon, a restless night, a forgotten appointment. We were inspired by a simple question: what if AI could turn those scattered everyday observations into something organized, explainable, and genuinely useful — without ever pretending to be a doctor?

What it does

NeuroLens AI is an AI-powered early-awareness and health-literacy platform for neurological and cognitive wellbeing. Users log everyday observations (focus, sleep, mood, headaches, daily functioning) into a visual timeline. A deterministic pattern engine scores how strongly those observations form a meaningful pattern worth monitoring, and an AI explanation layer translates the math into plain language: what was observed, what could influence it, what to keep tracking, and what questions to bring to a healthcare professional. It also includes 60-second daily check-ins, non-clinical cognitive mini-exercises, a shareable clinician-friendly health summary (PDF), a Learn section with sourced health education, and a Privacy Center with export and delete-my-data.

How we built it

Frontend: React + TypeScript + Vite + Tailwind CSS, with Recharts visualizations and Lucide icons — deployed on Vercel. Backend: FastAPI + Python with SQLAlchemy, JWT authentication, and PostgreSQL (SQLite fallback) — deployable to Render/Railway. AI pipeline: User input → validation → structured observation store → deterministic pattern engine → feature extraction → AI explanation layer (OpenAI-compatible provider abstraction, structured JSON) → safety validation → UI visualization. The LLM never diagnoses; it only explains pre-computed patterns, and every output is schema-validated and safety-filtered, with a deterministic Demo Intelligence Mode fallback so the app never breaks without an API key.

Challenges we ran into

The hardest problem was responsibility: building health AI that is genuinely useful without ever diagnosing. We solved it architecturally — the LLM is fenced behind a deterministic engine and a safety layer that blocks disease claims, certainty, and medication advice. We also had to make pattern math truly explainable (frequency, trend, persistence, severity, impact, 48-hour co-occurrence) and design a calm, non-frightening UX for sensitive topics.

Accomplishments that we're proud of

A complete, working product: real database, real auth, real pattern math, validated AI pipeline, PDF reports, safety guardrails, automated tests, seeded demo persona, live deployment — plus a one-page flow a judge can experience end-to-end in under 4 minutes.

What we learned

Longitudinal context beats one-shot answers; constraints (no diagnosis, explainability, privacy-by-design) made the product better, not smaller; and fallback-first design is what separates a demo that survives judging from one that doesn't.

What's next

Backend deployment with hosted Postgres, styled clinician PDFs, reminder nudges, multilingual plain-language summaries, and opt-in wearable sleep import.

Built With

• fastapi
• jwt
• lucide
• openai
• postgresql
• pytest
• python
• react
• recharts
• rest-api
• sqlalchemy
• tailwind-css
• typescript
• vercel
• vite