Browse categories
Explore
Fiverr Pro
English
$
USD
I will build ai agents, rag chatbots and llm applications in python
Said H
Level 1
About this gig
Most AI systems ship without anyone measuring whether they actually work. They pass a demo, then return the wrong passage on a real question, or an agent takes the wrong action.
I build yours and prove it works.
WHAT I BUILD
RAG systems over your documents. AI agents that use tools and take multi-step actions. LLM applications - extraction, classification, summarization, chat interfaces. Source code in every package.
WHAT MAKES THIS DIFFERENT
On Standard and Premium I build an evaluation set from your real questions and measure accuracy before delivery. You get numbers, not assurances.
ALSO
Already have a RAG system or agent that misbehaves? I diagnose why - chunking, embeddings, retrieval, ranking or tool selection - fix the dominant failure mode, and re-measure. Message me before ordering.
STACK
Python, LangChain, LangGraph, pgvector, Chroma, hybrid search, cross-encoder reranking, multilingual retrieval. AWS Bedrock, FastAPI, Docker. Claude, OpenAI, Hugging Face.
BACKGROUND
AI/ML engineer working on production AI systems. AWS Certified Machine Learning - Specialty.
Tell me what you are building, or what is breaking.
Get to know Said H
Said H
AI and ML Engineer, RAG and LLM Systems
Level 1
- FromMorocco
- Member sinceSep 2021
- Avg. response time1 hour
- Last delivery6 months
Languages
English, French, Arabic
AI/ML engineer. I build RAG and LLM systems, and I diagnose them when they don't work on real documents.
Most broken RAG systems aren't a model problem — the right passage never reaches the model, so no amount of prompt engineering fixes it. I measure where retrieval actually fails, fix the dominant failure mode, and re-measure so the improvement is visible rather than asserted.
Professionally I work on production RAG on AWS, speech recognition for low-resource languages, and model fine-tuning. AWS Certified Machine Learning – Specialty.
Tell me what's breaking.
My Portfolio
FAQ
What do you need from me to get started?
Your documents (PDF, Word, HTML, database, or a website to crawl) and a sense of the questions your users will actually ask. If you have real user questions or support tickets, send those — they make the system measurably better because I build the evaluation set from them.
How do you measure accuracy? What does that actually mean?
I build a set of questions with verified answers from your own documents, then measure how often the system retrieves the right passage. If that passage never reaches the model, no prompt engineering fixes the answer. You get the number before delivery.
My documents are not in English. Does that matter?
Yes, more than most people expect. Retrieval quality can differ substantially between languages on the same corpus with the same model. I work with multilingual retrieval regularly and measure per language rather than assuming one number covers both.
Can you fix my existing RAG system instead of building one?
Yes. Message me first so we can scope it — the price depends on whether the fix is chunking, embeddings, retrieval configuration, or something upstream in how documents were parsed. I diagnose before quoting.
Will it make things up?
Not silently. I build refusal behavior in, so when the answer is not in your documents the system says so rather than inventing one, and answers cite their source. That is a design requirement, not an afterthought.
Do I get the code?
Yes, in every package. You own what I build and can host, modify, and extend it yourself.
