Inspiration

We wanted to make medical information more accessible in a way that feels safe and non-judgmental. A public forum lets people share their lived experiences while still accessing reliable medical knowledge.

What it does

MedSource is an anonymous forum where people can share and read posts about medical topics—open, honest, and stigma-free.

How we built it

Front-end: Streamlit AI feedback: Groq + OpenFDA API using RAG (retrieval-augmented generation) Advanced search: Sentence-transformer embeddings for semantic search, not just keyword matches

Example: fetching posts semantically

def get_reviews(self, df, number: int, query: str):
      embeddings = np.vstack(df["embeddings"].to_numpy()).astype("float32")
      dim = embeddings.shape[1]
      faiss.normalize_L2(embeddings)
      index = faiss.IndexFlatIP(dim)
      index.add(embeddings)

      model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")

      #embed the query
      query_embedding = model.encode(query, convert_to_tensor=False)
      #format for the search 
      query_vector = np.array([query_embedding], dtype="float32")
      faiss.normalize_L2(query_vector)

      k = number  # Find the top 5 most similar items
      distances, indices = index.search(query_vector, k)

Challenges

Front-end development was tricky—we’re mostly back-end Python devs, and this was our first hackathon.

Accomplishments

Built a fully functional, scalable product in a short time Integrated AI feedback seamlessly with real-world APIs

What we learned

Web development intricacies AI + RAG implementation End-to-end product building under time pressure

What’s next

Extend MedSource with user profiles and richer experiences More forums, access points, and community-driven features

Built With

• faiss
• fda
• groq
• pandas
• python
• sentence-transformers
• streamlit