Service

LLM pipelines and AI agent development

LLM features that work on messy real input, from document pipelines to agents that take actions.

What's included

  • Document pipelines with OCR fallbacks for scanned files
  • LLM extraction and explanation with confidence scores
  • Agents that plan and act in a browser or an app
  • MCP tools and integrations
  • Async processing with live progress for users
See it in practice: FlexReport

I build LLM features that work on real documents. FlexReport explains 50 to 200+ medical terms per report and reads scanned PDFs as well as digital ones. ARIA, a browser agent I led the engineering on, won the Samsung Web Agent Hackathon.

How a project runs

We start from your real inputs, check quality on them as we go, and ship the pipeline behind a simple API.

Questions

Which models do you use?

Whichever fits the job and the budget. I have shipped with Llama 3.3 70B, Gemini 2.5 Flash and domain models like BioBERT.

Can it handle scanned PDFs?

Yes. I use a fallback chain, so text that one tool cannot read goes to the next.

How do we start?

Send a few sample inputs and what a good output looks like. I reply with what is realistic and a plan.

Start a project