I will build ai agents that automate your business workflows


About this gig
Most "AI agent" builds fail the same way: they work in a demo and break on the third real task. The difference is error handling, not the model.
I build agents that do a specific job end to end, know when they are stuck, and hand back to a human instead of guessing.
WHAT YOU GET
An agent scoped to a real workflow you run today, not a general assistant
Tool and API integrations so it can actually act, not just talk
Guardrails: retry logic, failure paths, and a human handoff when confidence drops
Logging you can audit, so you can see what it did and why
Deployed, documented, and maintainable by your team
STACK
Python, LangChain and LangGraph, OpenAI / Claude / open-source models, function calling, vector search, Docker, AWS.
WHY ME
Two master's degrees in AI (University of Luxembourg; DataScienceTech Institute Paris). Neo4j Certified Professional. I have shipped AI systems in healthcare and private banking, where an unhandled failure is not an option.
BEFORE YOU ORDER
Message me with the workflow you want automated and roughly how often it runs. Some tasks are cheaper to automate with a plain script, and I will say so.
Get to know Youssef.b
Data Scientist, RAG and LLM Engineer
- FromFrance
- Member sinceAug 2024
- Last delivery1 year
Languages
English, French, Arabic, Spanish
My Portfolio
FAQ
What kind of workflows can you automate?
Anything with a clear trigger, defined steps and a checkable result: processing inbound documents or emails, enriching and routing leads, monitoring and summarising data sources, generating recurring reports, or moving information between tools that do not integrate natively.
How is this different from Zapier or Make?
Zapier is excellent for fixed if-this-then-that rules. An agent is worth it when steps require judgement, the input is unstructured, or the path changes case by case. If your workflow is genuinely rule-based, I will tell you to use Zapier and save your money.
What if the agent makes a mistake?
That is a design question, not an accident. Every agent I build has explicit failure paths: it retries where retrying is safe, stops and escalates to a human where it is not, and logs every action so you can audit what happened. You decide where the human checkpoint sits.
Will it connect to the tools we already use?
If it has an API, generally yes. Common ones are straightforward: Slack, Notion, Google Workspace, HubSpot, Airtable, most CRMs and databases. Tell me which tools are involved before ordering and I will confirm what is reachable and what is not.
How much will it cost to run?
Model API usage, billed to your own account so you keep control. Cost depends on task volume and model choice, and I will estimate it before you order. Where a smaller or open-source model does the job as well, I use it, because it is usually far cheaper per run.
