Alexandre Beauvois

My role consists of shipping production-grade AI agents end to end, from the backend brain (nestJS) to the mobile experience (expo app). I design and build AIAgentService layers that wrap LLM providers (Anthropic, OpenAI, Google Gemini via LangChain) including:

- Design the agent loop: model call, tool dispatch, result feedback, stop conditions, max-iteration safety

- Pick and abstract the LLM provider (Anthropic / OpenAI / Bedrock) behind an interface so you can swap models or run A/B comparisons

- Define the tool/function schema, validation, and serialization layer (zod/drizzle)

- Build the prompt system: system prompts, few-shot examples, dynamic context injection, versioning

- Conversation/session state: message history, summarization when the context grows, memory persistence

- Streaming: SSE or WebSocket transport from the service to the Expo client, including partial tool-call streaming

- Retry, backoff, and graceful degradation when a provider rate-limits or errors mid-stream

- Cost and token accounting per user / per session (Ledger approach with idempotent side effects so agents never double-charge or double-send)

On the mobile side, I build Expo/RN client that solves Math problems based on a photo, with tool-call visualization, optimistic state, offline-aware persistence, push notifications for async agent completions, and deep links that let agents drive the app directly. I take prompt and agent versioning seriously, with eval harnesses, regression suites, and red-teaming against prompt injection gating every deploy. I also handle the work that keeps agents reliable: latency budgets (reduced loop after loop), prompt prefix caching, per-user quotas, abuse detection, and fallback model when primary down. I share strict TypeScript or zod schemas across backend, agent tools, and the Expo app so the contract is enforced everywhere. I bring the full stack of skills required to ship something users actually trust.

Metroscope created a B2B SaaS application which leverages AI to fight waste, pollution and uncertainty with monitoring & diagnostics features for Power Plants - Nuclear, CCGT - to optimize performance and cut downtime.

- As a full stack engineer I joined the Anomaly Detection feature team and delivered User Stories with a product manager, a UX Designer, 3 data scientists, 1 platform engineer and 3 other developers.

- The web tech stack: micro services/nats/NestJS/openapi/postgresql/prisma/Typescript/React/Azure Cloud/ (with complex charting features)

- The AI tech stack: micro services/nats/python/Jupyter (many models/training/predicting/deploying)

- The Infra stack: docker/kubernetes/terraform

- The Data tech stack: adx/airflow ( timeseries from hundred of sensors per plant every minute)

Metroscope created a B2B SaaS application which leverages AI to fight waste, pollution and uncertainty with monitoring & diagnostics features for Power Plants - Nuclear, CCGT - to optimize performance and cut downtime. - As a full stack engineer I joined the Anomaly Detection feature team and delivered User Stories with a product manager, a UX Designer, 3 data scientists, 1 platform engineer and 3 other developers. - The web tech stack: micro services/nats/NestJS/openapi/postgresql/prisma/Typescript/React/Azure Cloud/ (with complex charting features) - The AI tech stack: micro services/nats/python/Jupyter (many models/training/predicting/deploying) - The Infra stack: docker/kubernetes/terraform - The Data tech stack: adx/airflow ( timeseries from hundred of sensors per plant every minute)

Supervan is a kind of “Uber for heavy and large packages”. It offers last-mile transportation services and cutting-edge software solutions to leading companies such as IKEA, Weber, Leroy Merlin, and other major retailers, enabling them to optimize their supply chain operations, enhance their delivery capabilities, and drive greater efficiency and customer satisfaction. We handle 2000+ deliveries daily.

- Hands-on producing and supporting the 3 SaaS Apps in Typescript/React/postgresql

- Hired backend and frontend engineers for the new stack

- squad management of 3 devs + 1 PO + 1 UI Designer