- Build tool-using agents on LangGraph and LangChain, like GIS analysis agents for correlation, catchment and travel time.
- Protect confidential data with a Presidio + GLiNER guardrail: PII detection F1 from 0.58 to 0.91, Arabic at 1.00.
- Stop prompt injection in English, Egyptian Arabic and Arabizi, including hidden-character tricks.
- Evaluate LLM systems before users do: eval suites that catch bugs and measure safety across languages.