Symptom Timeline

landing

🧠 Inspiration

Every time we go to the doctor, we’re asked:

"When did your symptoms start?"

And every time, we struggle to remember the details. Especially for chronic issues, it’s hard to keep track of how things have changed over time. We wanted to build something simple but powerful: A natural language health journal that remembers for you — and gives doctors a clear view of your symptom history.

We also saw this as an opportunity to apply large language models (LLMs) in a truly helpful and responsible way — using them not just for chat, but for structured health insights that can genuinely improve care.

⚙️ What It Does

Symptom Timeline is a frontend web app that allows users to:

Write about their symptoms in plain English (like a diary)
Use LLMs to analyze and extract:
- Symptom names
- Start and end dates (temporal reasoning)
- Severity (optional)
View a clean, structured timeline of symptoms
Download or share the timeline with a doctor
(Bonus): Get a recommendation to seek medical attention if symptoms sound serious

🧰 How We Built It

Frontend: React + TailwindCSS (100% frontend-only)
LLM: OpenAI GPT-4 API for natural language processing
State: Used browser local storage (no backend required)
PDF Export: Used jsPDF to generate downloadable timeline summaries
Date Parsing: Prompt-engineered the LLM to recognize temporal cues ("two days ago", "last night", etc.)
UI: Designed with accessibility and simplicity in mind — friendly for all ages

💡 What We Learned

Prompt engineering matters — a lot. Getting the LLM to reliably extract symptoms and dates required many prompt iterations.
Users describe symptoms very differently. Some say "felt dizzy", others say "spinning room" — so we had to generalize and normalize.
LLMs are powerful at temporal reasoning, but need to be explicitly told to ground dates to a "current date" context.
Building a fully functional product with zero backend is totally possible with today’s tools — fast to prototype, and secure for users.

🚧 Challenges

Natural language ambiguity — not every sentence is clear or easy to structure. We had to tweak our prompts to handle vague phrasing.
LLM safety — we had to make sure the AI never gives medical advice, just reflects back structured summaries and flags when a doctor should be consulted.
Date handling — mapping human expressions like “a few days ago” into specific dates required custom logic and prompt tuning.