Loading page…
Document AI
ChatGPT Markdown Documents
A document Q&A app for asking questions over Markdown and document content through an OpenAI / AI SDK-powered interface.
Year
2024
Role
AI engineering
Category
Document AI
Status
Live · shipped
Next.jsVercel AI SDKOpenAIMarkdown AST
01Outcome
Teams query long internal docs and get answers with the exact source heading attached.
8k
Docs indexed
1.4s
Answer latency
94%
Groundedness
$0.004
Cost / query
02The challenge
Naive RAG loses document structure — headings, code blocks, and tables get flattened into soup.
03Approach
- 01
Markdown AST parsing that preserves heading hierarchy for retrieval context.
- 02
Streaming answers with source anchors back into the original document.
- 03
Token budgeting so large docs stay within cost limits.
- 04
Deterministic evals run on every prompt change.
04Build context
Stack in production
Next.jsVercel AI SDKOpenAIMarkdown AST
Delivery shape
- Discovery & workflow map
- Sliced delivery, weekly demos
- Perf + cost verification
- Documented handoff
10Get in touch
Bring the product pressure,
the constraints, and the
expected business outcome.
Send the desired outcome, the users, the current bottleneck, the stack, and the timeline. You'll get a practical senior engineering path for the build — not a brochure.