What I've built.
Real systems built to solve real problems — deployed, tested, and representative of how I approach production engineering.
DocAgent
An AI document processing agent that extracts structured data from any PDF, flags risks and discrepancies, and delivers plain-English summaries. Handles scanned documents via Tesseract OCR with multi-signal detection. Validates all output against Pydantic schemas and retries with progressively stricter prompts on malformed output. Every processing job logged with full status tracking.
↗ Live and deployed — supports invoices, contracts, legal filings, and scanned documents.
Arkive
An enterprise-grade knowledge base that lets teams upload documents and ask questions in plain English — returning cited answers grounded in the actual source files, with document preview and highlighted passages. Every query logged via Langfuse with full observability: latency, token usage, retrieved sources, and model response. Built for professional environments where accuracy must be verifiable.
↗ Live and deployed — supports PDF, DOCX, and TXT including structured tables.
NexaDesk
A white-label AI customer support widget that answers from a custom knowledge base, escalates to humans when genuinely needed (two-tier logic — resolves first, escalates only when required), and logs every conversation to PostgreSQL. Notifies support teams via Slack or email with full transcript on escalation. Deployable on any website with a single script tag.
↗ Live and deployed — includes a live brand customizer for instant white-label preview.
Have a problem that fits?
These systems represent the kind of work Refraxis takes on. If your workflow has a similar shape, get in touch.
Get Started →