À propos de Bernard
I'm a full-stack and AI engineer who ships production systems, not prototypes. Wiring an LLM to a database is easy — the hard part is making it trustworthy enough to put in front of your customers. That's exactly what I do.
My flagship project, Sentinel, is a live AI analyst that answers from 3,000+ real documents: hybrid retrieval with citations, a validated LLM-judge scoring every answer, and guardrails that refuse to guess rather than hallucinate — measured at 0% hallucination, ~$0.0075 per query. That's the standard I build to: reliability you can measure, not demos that fall apart in production.
I own the whole stack, from idea to deployed system:
- AI / LLM — RAG, evals, guardrails, Claude/Anthropic & open models
- Backend — Python, FastAPI, PostgreSQL / pgvector, Redis
- Frontend — Next.js, React, TypeScript, Tailwind
- Infra — Docker, Cloudflare, self-hosted Linux, 24/7 uptime
How I can help you:
- Add a production-grade AI feature to your app (RAG over your data, with citations + guardrails)
- Audit an existing LLM feature that's hallucinating or unreliable
- Build a full-stack MVP end-to-end, AI-enhanced where it adds real value
I hold an MSc in Computer Science (Cybersecurity) from EPITA Paris and recently completed an ML Engineer program. Based in Paris, working in English and French, invoicing via my company Devitify SAS.
Available now for remote and on-site projects — tell me what you're building and I'll tell you exactly how I'd ship it.
Anglais
Bilingue ou natif
Français
Notions
Accepte de travailler sur site
Paris (jusqu’à 50 km)
Expériences
- DevitifyFounder & Lead Engineer (AI / Full-Stack)AGENCE & SSIIjanvier 2026 - Aujourd'hui (6 mois)France
- Founded Devitify, an MVP & AI-agent studio for pre-seed founders and growth-stage teams; signed first paying client within 3 months of launch.
- Lead a 3-person engineering team — own architecture, technical hiring, code review, and delivery for every client engagement.
- Built the AI Agent Studio on Anthropic Claude + OpenAI Realtime (voice/text): SSE streaming, graceful provider fallback, multi-language response adaptation, and token-usage observability.
- Delivered a 16-week enterprise services platform for the first paying client — Next.js + NestJS microservices, 5 isolated PostgreSQL databases, 30+ Prisma models, Paystack + mobile-money + SMS integrations; Phase-1 MVP shipped in 9 weeks.
- Architected Devitify's internal platform end-to-end (admin console, client portal, marketing site, lead-capture microservice) — NestJS · Next.js 15 · React 19 · PostgreSQL 16 + pgvector · Redis · BullMQ · AWS S3.
- Implemented a security-first architecture aligned with GDPR, SOC 2, ISO 27001 — Argon2 + JWT service-to-service signing, granular RBAC, Cloudflare Turnstile, per-IP throttling, real-time WebSocket presence.
- EducityMachine Learning EngineerAGENCE & SSIIoctobre 2024 - décembre 2024 (2 mois)
- Integrated ML APIs into a VR-based adaptive-learning product — built algorithms that personalize content delivery based on student behavior patterns and learning velocity.
- Migrated ML inference onto containerized microservices with edge compute — cut model-update cycles from weeks to hours.
- Stood up internal MLOps tooling: automated data labeling, a model evaluation harness, model versioning, performance monitoring, and CI/CD pipelines for models.
- Owned technical documentation and onboarding runbooks for the ML team — standardized a workflow adopted across engineering.
- Phoebe SASCo-Founder & CTOAGENCE & SSIIoctobre 2024 - Aujourd'hui (1 an et 9 mois)Paris, France
- Co-founded Phoebe and own the product as sole architect/CTO — from whiteboard to live traffic in 12 months, setting the technical bar for every codebase.
- Built the platform end-to-end on NestJS 11 · Next.js 15 · PostgreSQL · Redis · Socket.io across 26 backend modules — real-time dashboards, leaderboards, multi-provider auth, and a transactional token ledger with atomic rollback.
- Designed the AI Aggregation Engine powering every forecast — a merit/entropy-weighted ML ensemble that compounds accurate forecasters and decays noise.
- Delivered a paid prediction study for a major AAA game studio (Amazon Games) ahead of a flagship launch: multi-phase structured deliberation, 100 participants across 20 groups, a 27-page predictive report — validated at a 1% error margin across 120 press reviews.
- Leading a live engagement for a European research institute — GDPR-first architecture, isolated PostgreSQL instances, AWS Frankfurt (CDK-managed), real-time deliberation via Socket.io + Redis.
Recommandations
Soyez le premier à recommander Bernard
Contribuez à la réussite de ce freelance en partageant votre expérience de collaboration avec lui.
Ces profils de freelance correspondent également à vos critères
Agatha Frydrych
Backend Java Software Engineer
4.7
(3)
2
Baptiste Duhen
Fullstack developer
4.6
(4)
5
Amed Hamou
Senior Lead Developer
4
(2)
7
Audrey Champion
Web developer
4.3
(3)
4
Formations
- MSc ComputerEPITA2023MSc Computer
- BScShenyang Institute of Engineering2021BSc