AI and automation
Models and workflows on a real task, including chat over your own knowledge.
What this work is
I connect a model to a task you already have. That includes a chatbot over your own knowledge (RAG), search across your documents, pulling fields out of a messy export, and agents that call an API you already trust and show what they changed.
A model only comes in when the task needs one. If a missing field, a filter, or a clearer screen would fix it, I'll say so. When a model does belong, the screen shows what it read, what it suggested, and what a person approved, including routing to local or open models when a paid call on every request is the wrong fit.
This covers
- Machine learning models
- Chat over your own knowledge (RAG)
- Assistants over your own documents
- Extraction and classification
- Orchestration and model routing
- Local and open models
- Tool calling and API automation
- MCP, where it connects existing tools
- Machine learning models
- Chat over your own knowledge (RAG)
- Assistants over your own documents
- Extraction and classification
- Orchestration and model routing
- Local and open models
- Tool calling and API automation
- MCP, where it connects existing tools
- Machine learning models
- Chat over your own knowledge (RAG)
- Assistants over your own documents
- Extraction and classification
- Orchestration and model routing
- Local and open models
- Tool calling and API automation
- MCP, where it connects existing tools
- Machine learning models
- Chat over your own knowledge (RAG)
- Assistants over your own documents
- Extraction and classification
- Orchestration and model routing
- Local and open models
- Tool calling and API automation
- MCP, where it connects existing tools
- Machine learning models
- Chat over your own knowledge (RAG)
- Assistants over your own documents
- Extraction and classification
- Orchestration and model routing
- Local and open models
- Tool calling and API automation
- MCP, where it connects existing tools
When it fits
A repeated task that should be shorter, with a person still able to check the result.
You leave with
- Less manual work on a specific task, with a person still able to see what the system did.
- A repeated task removed or shortened
- Answers grounded in your data
- A trail of what the system changed
Where this shows up
Some of the projects are written up on the projects page. Writing on this kind of work is in Give an agent a boundary people can see.
The other three
Start on AI and automation.
Say what you need this to do, who it is for, and what already exists. I reply by email, and we agree scope in writing before work starts.