AI / Full-Stack Engineer — Jerusalem, Israel · open to Tel Aviv / hybrid

I build LLM-powered products end to end: the model layer, the backend behind it, and the native clients on top. Seven years of production experience, the last two spent making AI features reliable enough to ship — not just impressive in a demo.

Leirod — an AI nutrition and coaching app, live on the App Store and Google Play. Designed, built and released solo across four production codebases: a Hono/Node.js API, native iOS (Swift/SwiftUI) and Android (Kotlin/Jetpack Compose) clients, and a Next.js site. ~1,100 commits, 1,500+ automated tests.

Some of the engineering I'm most happy with:

• Multi-provider LLM routing with real failover. Every AI feature declares a primary → fallback → tertiary chain across Gemini, Claude, GPT and Mistral. Model selection is config-driven — swapping a model is an environment variable, never a deploy. A provider outage degrades the feature instead of breaking the product. Google models are pinned to Vertex AI so EU data residency holds.
• Vision-based food recognition. A vision model extracts foods and portions from a photo; the result is matched against the USDA FNDDS and French CIQUAL nutrition databases in PostgreSQL. Full nutritional breakdown in about two seconds per scan.
• A streaming chat agent with tool use. Provider-agnostic multi-turn tool loop operating on the user's own data — log a meal, correct a portion, query history. Per-turn credit billing is protected by HMAC-signed, user-bound, short-lived tokens, so billing can never be bypassed from the client.
• Deterministic scoring, narrative LLM. Weekly health scores are computed in code, not by a model. The LLM only writes the prose around them. Scores stay reproducible and auditable — a model is the wrong tool for arithmetic anyone might contest.
• Production ops I own. GitHub Actions → Linux VPS deploys behind a health-check gate, database migrations with fail-fast backups, encrypted off-site backups, scheduled workers, rate limiting, OAuth 2.0 (Google / Apple), Sentry and PostHog.

I also deliver LLM agents to small businesses — Telegram-based assistants automating back-office and operations workflows, currently saving three clients around ten hours a week each.

AI/LLM — agents & tool use, RAG, vision models, streaming, evaluation, fallback & cost control · Gemini, Claude, GPT, Mistral, OpenRouter, Vertex AI Languages — TypeScript, JavaScript, Python, Swift, Kotlin, SQL, PHP Backend — Node.js, Hono, PostgreSQL, Drizzle ORM, MongoDB, MySQL/MariaDB Frontend — React, Next.js, Vue, Tailwind CSS Mobile — Swift/SwiftUI, Kotlin/Jetpack Compose, App Store & Play Store releases Infrastructure — Docker, Linux, GitHub Actions, Nginx, Cloudflare, pm2

• Portfolio — doriel.dev
• LinkedIn — linkedin.com/in/leirod
• Email — doriel.chiche@gmail.com

Trained at École 42 (2016–2020). French (native), English (fluent), Hebrew (professional working proficiency).

The archived repositories below are coursework from my time at 42 — kept for the record. Current work lives in private product repositories; happy to walk through any of it in a conversation.