I build AI that cites its sources or says it doesn't know.

Reads Arabic and English tenders and tells a bid manager whether the bid can actually be submitted. It refuses to call a requirement met unless an approved, cited, in-date document proves it.

• Every requirement is a verbatim quote with its page; nothing reaches "satisfied" without approved evidence.
• 1,309 of 1,528 production requirements quoted verbatim on the cited page.
• Arabic and English tenders, PDF or DOCX, with selective OCR.

An Arabic-first platform for independent academies. Each academy gets its own storefront, sells courses, teaches through lessons, assessments and live sessions, and takes payment into its own account.

• Multi-tenant isolation enforced in the database with deny-by-default row-level security.
• Each academy is merchant of record on its own Stripe or Paymob account; the platform never holds funds.
• Lesson Q&A answers only from the transcript, with verbatim quotes, or refuses.

A clinical assistant that turns an Egyptian-Arabic doctor–patient consultation into live differential diagnoses, editable SOAP notes, prescriptions, lab orders and medication-safety checks.

• Streaming speech-to-text across four engines with automatic fallback.
• LLM safety findings are filtered against the reconciled medication list before a doctor sees them.
• Approved by the clinic owner as the platform going forward.

Multi-agent code review that runs in your own CI with your own LLM key, and shows which agent found what. Published on the GitHub Marketplace, with its real-world recall published too.

• Four specialist analysts, three samples each, majority vote and a verifier before anything posts.
• Findings whose quoted evidence isn't in the diff are dropped, mechanically.
• 28% recall on 60 real bug-fix PRs, published beside the 91% seeded score.

Privacy-first English/Arabic agentic RAG where access control, citations and privacy are enforced by the system, not requested in a prompt.

• Role-based access enforced inside the vector-database filter, for dense and sparse search alike.
• Hybrid retrieval fused by reciprocal rank, then cross-encoder reranking.
• Corrective re-retrieval that says "not found" rather than guessing.