# Sadok Bouzayen — Forward Deployed Engineer · AI Systems

> Sadok Bouzayen is a Forward Deployed Engineer (FDE) and AI systems engineer based in Tunis, Tunisia (CET). He ships AI-native products end to end — agents, evaluation harnesses, model routing, local inference, developer tools — and is available remote or on contract for AI-native startups in Europe and North America. Funded founder of Bookeeni through RedStart Tunis. Research paper currently under peer review. $1M+ enterprise sales record. Speaks Arabic, French and English.

## Contact
- Email: tunisstarter@gmail.com
- LinkedIn: https://linkedin.com/in/sadok-bouzayen
- Response time: under 24 hours

## Current
- Forward Deployed Engineer at an AI-native startup (2026 — now)
- Founder & Engineer, Bookeeni — backed by RedStart Tunis (2025 — now)
- Research paper under peer review (2026)

## How he builds with models
1. Deterministic layer under the model: parse, normalize, validate before inference.
2. Measure before fine-tuning: dense-context injection beat a fine-tuned build on quality and latency (GHelper).
3. Private data stays on the machine: client-side processing, in-browser embeddings (CodeBaseTXT, MKTCH).
4. Every external call has an explicit fallback (Bookeeni Radio, Bookeeni demo mode).
5. Tests and ADRs even solo: 33+ verification tests on Organizer OS.
6. Owns the whole loop: architecture, data, security (RLS, token hashing), deploy, on-call.

## Selected work
- Organizer OS (internal): agentic workspace generated from an event brief; deterministic multilingual parsing (FR/EN/AR/Derja), model routing via NIM proxy (Kimi, Nemotron), offline Whisper, 33+ verification tests, ADRs.
- MKTCH (internal): privacy-first matching engine on 384-d multilingual embeddings with in-browser inference; bilateral matching with hard constraints and ranking explanations.
- GHelper (internal): consultative AI sales assistant over a normalized product catalog; dense-context injection outperformed fine-tuning.
- CodeBaseTXT: https://codebasetxt.vercel.app — repository to LLM-ready context, fully client-side, ~30% token savings via Skeleton Mode.
- CodeViz: https://code-viz-nine.vercel.app — client-side dependency graphs via the TypeScript Compiler API, with a Gemini reasoning layer.
- Free AI Diagram Generator: https://freeaidiagramgenerator.vercel.app — text to the right diagram type, D3 rendering.
- Bookeeni: https://bookeeni.com — artist-first booking platform (React, TypeScript, Supabase, Postgres).
- ZYAT: https://zyat.bookeeni.com — live event platform with passwordless identity, Supabase RLS, serverless API, SHA-256 token hashing.
- Torque: https://torque.bookeeni.com — festival ticketing PWA with web push.
- Bookeeni Radio / The Orrery: https://radio.bookeeni.com — live scientific data sonified with Web Audio and a custom Canvas 3D renderer.

## Stack
TypeScript, React, Python, Supabase/Postgres, LLM APIs (Gemini, Claude, OpenAI, NVIDIA NIM), Whisper, embeddings and local inference, Web Audio/Canvas, Vercel/Cloudflare serverless.

## FAQ
- What does a Forward Deployed Engineer do for an AI-native startup? Scopes the use case, integrates the LLM into real customer data and workflows, builds the evals that prove it works, stays on the hook after launch.
- Remote or contract? Both. Tunis (CET), Europe and North America hours.