SubSearch

semantic search engine for reddit. type a concept, get back relevant posts — no keyword matching.

---

how it works

scraping — hits the reddit API for a given subreddit, pulls recent posts + metadata.

vectorizing — a small docker sidecar we built (./embedding-service): FastAPI + all-MiniLM-L6-v2. hit POST /embed with text, get back a vector. runs on port 8001, starts before the backend.

storage — embeddings go into ChromaDB, a lightweight vector database running in its own container.

search — your query gets embedded the same way, then ChromaDB finds the closest matches by cosine similarity.

backend — Express server wires it all together. exposes endpoints to index a subreddit and search across it.

frontend — Next.js UI, dark theme, nothing fancy. just a search box that hits the backend.

---

stack

• chromadb — vector store, port 8000
• embedding-service — sentence-transformers (all-MiniLM-L6-v2), port 8001
• backend — Express/TypeScript, port 3004
• ui — Next.js 16 + Tailwind v4, port 3000

---

running it

docker-compose up

everything starts together. hit localhost:3000 when it's up.

---

indexing a subreddit

GET /index?subreddit=sideprojects&limit=100

then search:

POST /search
{ "query": "indie hackers building in public", "limit": 10 }