Dx

[Devpost Link]

A simulated patient diagnosis platform for medical students.

Dx gives students a place to practice patient conversations, make a diagnosis, and learn from the clues they gathered. Each case pairs a fictional patient with structured disease and symptom data, then turns the encounter into a practical review of clinical reasoning.

At a glance

1. Meet a patient. Start a simulation with a fictional patient profile and a hidden diagnosis.
2. Have a conversation. Ask questions and listen to the patient through an ElevenLabs conversational agent.
3. Make a diagnosis. Search the condition library and submit your assessment.
4. Review the case. Compare your answer with the case diagnosis and see which symptoms overlapped or pointed elsewhere.
5. Spot patterns. After five completed cases, Learning Insights highlights recurring diagnosis mix-ups and symptoms to revisit.

Inspiration

We built Dx to give medical students a place to practice the human side of diagnosis: speaking with patients, understanding their stories, and connecting the details. A simulation lets students work through those conversations without real-world consequences for a patient.

What it does

Dx lets students speak with a simulated AI patient, gather clues, and submit a diagnosis. Afterward, they can see which symptoms supported their answer, which clues were linked to a different condition, and what they could review before the next case.

The Learning Insights page looks across completed simulations to show conditions a student has confused and symptoms that appeared in those cases. It becomes available after five completed assessments, when there is enough case history to make the review useful.

How we built it

Part	Technology	Role
Frontend	Next.js	Dashboard, patient encounters, case reviews, and learning insights
Accounts	Auth0	Sign-in and account-specific simulation history
Patient conversations	ElevenLabs Conversational AI	Voice interaction with the simulated patient
Case generation	Gemini	Turns ranked disease and symptom associations into structured fictional patient cases
Backend	Python	Case generation, simulation history, diagnosis reviews, and learning insights
Persistent storage	Turso	Stores generated cases and simulation history across serverless runs

The disease associations come from the public Disease-Symptom Knowledge Database from Columbia University Biomedical Informatics. The source ranks associations; those ranks are not symptom probabilities. Dx uses the data to build educational cases and compare the clues selected for each encounter.

Challenges we ran into

• Making patients feel distinct. We wanted the same condition to appear with different identities and backgrounds while keeping its clinical clues consistent.
• Protecting the reveal. The patient conversation needed to feel natural without exposing the diagnosis before the student submitted an answer.
• Connecting voice to a specific case. Each ElevenLabs conversation needed the right patient context and opening line.
• Keeping history reliable. Local SQLite data disappeared when the Vercel backend ran on temporary serverless storage, so we moved persistent data to Turso.
• Working with the source data. Parsing the public dataset and preserving its original association order took more care than we expected.

Accomplishments that we're proud of

Students can sign in, meet a patient, have a conversation, make a diagnosis, and receive feedback in one flow. The same disease can return with a new patient identity, so repeated practice does not feel identical. We are also proud of the interface, which makes the experience approachable while keeping the learning review useful, and of getting ElevenLabs conversational agents working with individual cases.

What we learned

A useful simulation needs more than a correct answer. The patient has to respond naturally, and the review has to help students understand why they made their choice. We also learned how important persistent storage is when an app runs on serverless infrastructure.

What's next for Dx - Simulated Patient Diagnosis Platform

We want to improve voice reliability, offer deeper feedback, and help students track how their clinical reasoning develops over time. We also want to explore richer medical datasets and responsibly sourced public patient narratives to make future cases more varied and realistic.

Run locally

The repository has a Python backend at the root and a Next.js app in frontend/. The source disease database is included in data/.

Backend

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
python app.py

Set GEMINI_API_KEY and DX_BACKEND_TOKEN in the root .env. For local development, Turso credentials can be left blank; the backend will use local SQLite. The backend runs at http://127.0.0.1:8000 by default.

Frontend

cd frontend
npm install
cp .env.example .env.local
npm run dev

In frontend/.env.local, set your Auth0 credentials, ELEVENLABS_AGENT_ID, DX_BACKEND_URL=http://127.0.0.1:8000, and the same DX_BACKEND_TOKEN used by the backend. Open http://localhost:3000 to use the app.