Browse categories
Explore
Fiverr Pro
English
$
USD
I will fix and improve your rag chatbot accuracy
Amine E.
About this gig
Is your RAG chatbot giving wrong answers, missing the right evidence, or sounding confident when the source does not support it?
I systematically test the path from question -> retrieval -> context -> answer -> citation.
I first check what your system can actually measure. Then I build a baseline, identify the first failing stage, apply the agreed improvement, and rerun the same test cases.
You can receive:
- evaluation-readiness check and baseline scorecard
- retrieval, grounding, citation and follow-up diagnosis
- scoped code/config fixes
- before/after results and regression analysis
- rollback guidance and technical handoff
In an independent 50-case SourceChat practice audit, 39/50 final answers were correct. On the same 45 answerable cases, fully correct answers improved from 26/45 to 34/45 and exact-ID retrieval reached 8/8 without rebuilding the existing 10,736-chunk index.
Python/FastAPI, Node.js/TypeScript, LangChain/custom RAG, Pinecone, pgvector, OpenAI and Gemini.
Message me before ordering with your stack and 3 failing examples.
Get to know Amine E.
Amine E.
RAG and Full Stack AI Developer
- FromMorocco
- Member sinceAug 2026
Languages
Arabic, English, French
I build grounded AI assistants and RAG chatbots that turn PDFs, manuals, SOPs, policies, and internal knowledge into reliable answers with source citations.
I focus on retrieval quality, document ingestion, exact identifiers, insufficient-evidence handling, and clean web chat experiences.
I’ve built SourceChat, a working multi-format document RAG app, plus a technical knowledge assistant using hybrid vector + full-text retrieval.
My stack includes Node.js/NestJS, React, OpenAI, Gemini, PostgreSQL/pgvector, Pinecone, and LangChain.
My Portfolio
FAQ
Do I need an existing RAG chatbot?
Yes. This Gig is primarily for diagnosing and improving an existing RAG chatbot or retrieval pipeline. If you need a new RAG assistant built from scratch, please use one of my RAG development Gigs.
Can you guarantee zero hallucinations?
No responsible RAG system can guarantee that an LLM will never produce an incorrect response. I focus on improving retrieval, grounding, citations and insufficient-evidence behavior, and on measuring the system with test questions.
What RAG problems can you investigate?
Examples include irrelevant retrieval, missing documents, weak chunking, metadata filters, exact identifiers not being found, poor context construction, citation problems and unsupported answers.
What technologies do you work with?
My main stack is Node.js/TypeScript, LangChain, PostgreSQL/pgvector and OpenAI. I can also review custom RAG architectures where the code and components are accessible.
What do you need from me?
I need a description of the problem, your current architecture or repository, sample documents/data if required, and examples of questions where the system performs incorrectly.
Will you measure improvement?
Standard and Premium can include an evaluation set and before/after testing so improvements are based on actual test cases rather than subjective impressions.
Can you implement hybrid search?
Yes, when it is appropriate for the problem. This may combine semantic retrieval with keyword or full-text retrieval, particularly when users need both natural-language search and exact identifiers.
