Inspiration
Many people search for symptoms online, often receiving misleading or overwhelming results. We wanted to build a responsible, AI-powered assistant that retrieves medically relevant cases, gives structured insights, and empowers people with clear, actionable next steps.
⸻
What it does
MediChain AI analyzes user-entered symptoms, finds similar cases with TiDB Serverless vector search, suggests possible health conditions with a simple reasoning layer, and provides recommendations — including whether to monitor at home, see a doctor, or visit a nearby clinic.
⸻
How we built it • Collected and prepared a small dataset of symptoms and associated conditions. • Used Python to generate embeddings and ingest data into TiDB Cloud Serverless. • Enabled TiDB Vector Search to retrieve the most relevant medical cases for a given symptom query. • Built a reasoning layer in Python to assign probabilities and compute a risk score. • Designed a lightweight CLI/Notebook demo that takes symptoms and outputs conditions + recommendations.
⸻
Challenges we ran into • Ensuring the dataset was clean, consistent, and medically relevant. • Getting vector search tuned for small data without noisy matches. • Balancing simplicity with usability — making the system informative while not over-promising as medical advice. • Orchestrating embeddings, queries, and reasoning within a minimal stack.
⸻
Accomplishments that we’re proud of • Built a fully working multi-step agentic pipeline powered by TiDB Cloud. • Delivered a demo that is simple to run yet shows a real-world healthcare use case. • Created an intuitive workflow: from symptoms → retrieval → reasoning → recommendation. • Kept the system responsible with disclaimers and risk scoring.
⸻
What we learned • How to leverage TiDB Cloud Serverless for vector search and structured data together. • The importance of dataset design in medical AI projects. • How chaining multiple steps (retrieval + reasoning + recommendation) creates a strong agentic experience. • That small, well-scoped demos can still show big impact.
⸻
What’s next for MediChain AI (Smart Symptom Checker) • Add multilingual support (English, Korean, Kazakh, etc) to broaden accessibility. • Integrate wearable data (heart rate, sleep, steps) for richer insights. • Expand dataset to include more symptom–condition mappings and medical guidelines. • Build a web or mobile interface for easier use beyond notebooks. • Incorporate clinic finder APIs globally for personalized recommendations.
Log in or sign up for Devpost to join the conversation.