I will build a production ai agent backend with rag and tool calling


About this gig
I build agent systems that hold up when the model misbehaves - not chatbot demos.
Relevant work:
- Led the backend of an ERC-8004 research-agent platform: users create, configure and run research agents, with automated datasource reconciliation, parameter normalization, code generation and a run-test evaluation loop. Agents run on Cloudflare Workers and Containers.
- Built an LLM-driven derivatives trading system on an Agent/Worker architecture - LangChain for strategy decisions, workers for execution. A user trading through it took first place in the WEEX AI Trading competition finals.
What you get:
- An agent with real tool calling (not just function stubs), plus retries and failure handling
- Retrieval over your own documents when the agent needs to know things
- An evaluation loop so you can change a prompt and actually know if it got better or worse
- Clean TypeScript or Python, with tests
I have been building with LLMs since 2024 and work fully AI-assisted myself, so I know where agents break: tool schemas, state, retries, context bloat.
Tell me the task your agent must do and where it is failing now - I will tell you if it is fixable or needs a rebuild.
Get to know Gowinder
nestjs, jsredis, postgres, rabbitmq, microservices, python, golang
- FromChina
- Member sinceAug 2024
Languages
English, Chinese
My Portfolio
FAQ
Can you fix our existing agent instead of rebuilding?
Usually yes. Send me the repo and the failure case and I will tell you honestly which is cheaper.
Whose cloud account does it run in?
Yours. I work in your account or hand you a reproducible deploy script.
Do you use our prompts or write new ones?
Either. If you have prompts that mostly work, I keep them and build the reliability layer around them.

