Browse categories
Explore
Fiverr Pro
English
$
USD
I will build document search or a rag API with source references
Nik Nastev
About this gig
Make a small set of business documents searchable, with references back to the source.
I build a single-user, single-corpus document retrieval service using Python and Postgres/pgvector or your existing compatible Supabase. Basic delivers document search. Standard and Premium add a question-answering API using one agreed model provider, source references and a fallback when evidence is insufficient.
You receive source code, ingestion instructions, tests and a handover guide. PDF inputs must contain selectable text; TXT and Markdown are also supported. No scans, OCR, spreadsheet parsing or website crawling.
The offer covers a bounded retrieval system, not a general chatbot, long-term conversation memory or model training. A chat UI, enterprise permissions and ongoing hosting are separate. Answers can still be wrong; citations and sample checks help review them. Client-funded infrastructure, embeddings and model usage are required where used.
Get to know Nik Nastev
Nik Nastev
AI Agent Orchestrator, MCP, RAG and Automation
- FromBulgaria
- Member sinceOct 2026
Languages
English, Bulgarian, Russian
I'm Nik Nastev, founder of ArtminAI. I design AI agent workflows and build integrations using TypeScript/Node.js, REST APIs, MCP and RAG. My work includes AI-native CMS, persistent vector memory and multi-agent development with separate coding, QA and deployment lanes. I use AI-assisted development with architecture, code review and verification gates. Available for scoped remote projects, integrations and workflow reviews.
Other AI Development Services I Offer
FAQ
Is this persistent conversation memory or a chatbot?
It is persistent storage and retrieval of the agreed documents. Standard/Premium add a Q&A API. Conversation history, a chat interface and autonomous memory extraction are not included.
Which documents can I use?
Selectable-text PDFs, TXT and Markdown within the file, page and size caps. One A4-equivalent page is up to 500 words for plain text. You must have permission to process the material; scanned files and OCR are excluded.
Can you guarantee every answer is correct?
No. I include source references, insufficient-evidence handling for generated answers and a small agreed evaluation set. A person should review important outputs. We agree which provider may process your documents before ordering.
