Stop searching your data.Start understanding it.
QueryPilot searches across your documents, brings scattered pieces of evidence together, and shows you exactly where the answer comes from.
Plan first. Retrieve second. Answer with evidence.
Every question is planned before anything is searched, and every answer ships with the sources behind it.
The Planner
Reads the question, works out what it actually refers to, and decides where the answer has to come from: a document, a table, memory, or a mix. Nothing gets searched until the plan is built.
- Routes to documents, tables, or both
- Turns “last month” into real date bounds
- Every answer ships with its sources
Understood as
Read from
How QueryPilot is built.Open any of the five to study it in detail.
From local development to production. Containers, DNS and HTTPS, a reverse proxy with rate limiting, and CI/CD, from a local repository to a running cloud app.
Deployment Stories
Docker / CI/CD / HTTPS
From raw files to searchable knowledge. Parsing, extraction validation, structure-aware chunking, token checks, embedding and indexing, with retries when a stage fails.
Document Ingestion Pipeline
Chunking / Embeddings / Recovery
Process expensive tasks without blocking the API. Slow work runs outside the request cycle on a pool of workers, with retries, backoff, a dead-letter queue and idempotency keys.
Background Jobs & Concurrency
Workers / Retries / Idempotency
Keep the data layer reliable. MySQL holds the truth, ChromaDB holds vectors, Redis holds temporary state. Nothing is left orphaned.
Database Architecture & Consistency
MySQL / Redis / ChromaDB
From login to an authenticated request. JWT access and refresh tokens, expiry, refresh and logout, Google OAuth and the auth middleware.
Authentication & Token Lifecycle
JWT / Refresh / Google OAuth
On questions that span several sources, QueryPilot keeps answer quality climbing as hybrid retrieval, reranking and the planner agent each add to it, while standard retrieval flattens out. The chart is illustrative.