← Back to Portfolio RAG Pipeline

0.1% DEV — RAG Pipeline

21 steps · 4 phases · Startup Indexing → Query → Re-ranking → Answer Generation

embed("Why does npm link matter?")  →  cosine search → lexical re-rank → GPT-4.1-mini
—
Press Play or → to begin
Transcripts chunked → embedded → indexed at startup. Queries embedded in same space → cosine similarity → lexical re-rank → GPT-4.1-mini generates grounded answers with source citations.
⚛️
React SPA
Frontend
⚙️
RagService
FastAPI
✂️
Chunker
180w / 35w overlap
🗄️
VectorStore
In-memory cosine
🤖
OpenAI
text-emb-3-small
✨
Generator
gpt-4.1-mini
✨ Grounded answer delivered with lecture citations
Step — of 21
Press Play or → to begin
Transcripts chunked at startup → embeddings indexed → query embedded in same space → cosine search → lexical re-rank → threshold gate → GPT-4.1-mini generates answer — sources cite exact lecture and timestamp.
Data / Command
—
Speed 1.5×