// PROJECT_ATLAS_INITIAL_COMMIT
stop command-f'ing your pdfs.
Upload PDFs, ask questions in plain language. Built for teams who live in specs, reports, and long-form knowledge—not search boxes.
peeking under the hood
NEXT.JS 16
SUPABASE (PGVECTOR)
GROQ + LLAMA 3 + SCOUT
UNPDF PARSER
the flow
- 01
[ INPUT ]
pdf document
Upload one or more PDFs; we preserve structure for faithful retrieval.
- 02
[ PARSE ]
unpdf parser
Extract text and layout so chunks stay aligned with the source page.
- 03
[ CHUNK ]
split & window
Semantic splits sized for context windows and cleaner embeddings.
- 04
[ EMBED ]
transformers
Vectorize passages so similarity search finds the right evidence.
- 05
[ STORE ]
pgvector
Persist vectors in Postgres with Supabase for durable knowledge.
- 06
[ QUERY ]
groq + llm
Ask in natural language; answers cite the spans we retrieved.
// LOGIC FLOW UTILIZES ASYNC/SYNC RETRIEVAL TO MINIMIZE LATENCY AND MAXIMIZE PRECISION.