Browse categories
Explore
Fiverr Pro
English
$
USD
I will fix your rag pipeline or langgraph ai agent
Atta Ur Rehman
About this gig
Your RAG pipeline or AI agent worked in testing but gives wrong answers in real use? I find the root cause and fix it.
I build production RAG and multi-agent systems in regulated financial services, where wrong answers and data leaks aren't acceptable.
Problems I fix:
- Retrieval that misses the right passage (chunking, embeddings, metadata filters)
- Hallucinations and answers without sources
- LangGraph or LangChain agents that loop, skip tools or lose state
- Vector database issues in Qdrant, FAISS or Chroma
- Slow or expensive pipelines
How I work:
1. Reproduce the problem with your code, logs and failing questions
2. Find the root cause and agree the fix with you
3. Fix it and prove it on a before/after test set of your own questions
4. Document every change
Stack: Python, LangChain, LangGraph, Qdrant, FAISS, FastAPI, Docker, Ollama, OpenAI, Claude. I can also debug n8n workflows and MCP tool connections around your agent.
Not sure which package fits? Message me your stack and an example of a bad answer before ordering.
Get to know Atta Ur Rehman
Atta Ur Rehman
AI engineer and solution architect for RAG and AI agents
- FromPakistan
- Member sinceJun 2020
- Avg. response time1 hour
- Last delivery6 days
Languages
Urdu, English, Hindi
AI engineer building private RAG systems, AI agents and AI voice receptionists that work in real use, not just in demos.
By day I build AI systems for a regulated bank on private infrastructure, so data privacy, guardrails and audit logging are my defaults. As a freelancer I've built a fully offline document research assistant for an enterprise client and delivered a code audit and recovery plan for an AI business platform.
Stack: Python, LangGraph, LangChain, Qdrant, FAISS, FastAPI, Docker, Ollama, Vapi, Retell, n8n.
Tell me the problem and I'll tell you honestly if AI is the right fix.
My Portfolio
Other AI Development Services I Offer
FAQ
What do you need from me to start?
Repo access (read-only is fine) or a zip, how to run it, a few questions or inputs that fail, and what you expected instead. Logs help a lot.
Which stacks do you work with?
Python with LangChain, LangGraph or CrewAI; Qdrant, FAISS or Chroma; FastAPI and Docker; OpenAI, Claude or local models with Ollama. For anything else, message me first.
What if there are more than 2 problems?
The Diagnosis tells you exactly what is wrong and what each fix involves. If the work goes beyond your package, I send a custom offer before doing anything extra.
Can you work with sensitive data?
Yes. I prefer read-only access, touch only what the task needs, and hand back or delete copies at delivery. For fully private setups I can run everything locally with Ollama.
Do you guarantee an accuracy number?
I measure before and after on an agreed set of your own questions, so you see the improvement. I don't promise a fixed accuracy figure before I've seen your data.
