01
Document ingestion
PDF uploads become normalized page text and chunk metadata for retrieval.
10MB / PDF
Kimi chat / DeepSeek retrieval / Supabase pgvector
Chat with PDFs.
Get cited answers.
Stack
Next.js 16 / Tailwind v4 / TypeScript
Retrieval
DeepSeek V4 Pro embeddings
Answers
Source-cited chat responses
Upload internal documents, retrieve the right context, and ask grounded questions without guessing.
Workflow
From uploaded PDF to cited answer.
01
Upload
Extract clean page-level text
02
Retrieve
Find the most relevant chunks
03
Answer
Generate cited responses
04
Compare
Route between model outputs if needed
System
A lean RAG product, not a chatbot wrapper.
02
Vector retrieval
DeepSeek-compatible embeddings are stored and queried through Supabase pgvector.
5 chunks
03
Grounded answers
Kimi receives retrieved context and cites factual claims back to source chunks.
cited by page
04
Model routing
Compare endpoint runs both models over the same retrieved context.
2 models
Live demo
A square-edged workspace for document questions.
Q4 Research - PDF Workspace
Kimi chat / DeepSeek retrieval / pgvector context
Active
Documents
00
Chunks
05
Answers
00
Citations
00
Upload and select a document to start chatting.
Business case
Fast answers with visible sources.
The value is less time searching, fewer unsupported answers, and a clear path from prototype to internal AI workflow.
Researcher reading docs
$30/hr
Manual follow-up
Notion AI seats
$20/user/mo
Workspace-bound
DocChat API demo
~$5/mo
Document Q&A
Built by Anil
Want this for your team's docs?
I design, build, and automate full-stack AI products for SMBs and agencies.