Skip to content
Nicolas Benoit

Backend & systems engineer · France

Nicolas Benoit

I'm a developer who wants to know what's under every abstraction I use. So I rebuild them. and more

View projectsGet in touch

experience

  1. 2023.03 → present

    Infofil · Software developer, Back-end & Systems

    The work runs from web back-ends to real-time systems, and some of it ships to air-gapped networks that never touch the internet. The projects below have the detail.

    • Python
    • Go
    • FastAPI
    • NestJS
    • Next.js
    • PostgreSQL
    • WebRTC
    • OPC UA
    • Docker

projects

  1. Real-time video platform

    Low-latency, multi-client video streaming with live sensor data, deployed on air-gapped infrastructure.

    context

    At Infofil. NestJS. Designed and deployed to production.

    challenge

    Low-latency WebRTC broadcast of five camera feeds through a mediasoup SFU, sensor acquisition from an OPC UA server via a dedicated microservice, FFmpeg recording and a configurable live dashboard.

    decisions

    • A mediasoup SFU for multi-client fan-out, with a dedicated microservice bridging the OPC UA sensor source.
    • NestJS
    • TypeScript
    • WebRTC
    • mediasoup
    • OPC UA
    • FFmpeg

  2. IT support dashboard

    Internal ticketing and support tool used daily by the support team, replacing scattered tracking with a single system.

    context

    At Infofil. Go and Python/FastAPI back-end, Next.js front, used daily by the seven-person support team.

    challenge

    Full-text search over hundreds of archived documents, SLA tracking and task/project management, on PostgreSQL with Synology NAS integration over SFTP.

    decisions

    • Full-text search built on Bleve (Go) rather than standing up a separate search service.
    • Go
    • Python
    • FastAPI
    • Next.js
    • PostgreSQL
    • Bleve

  3. N1 support chatbot

    A first-line support chatbot backed by a local ML pipeline and a self-hosted LLM.

    context

    At Infofil. End-to-end pipeline, integrated into a Next.js corporate site.

    challenge

    Zero-shot intent classification without labelled data: generated a ~1,600-example synthetic dataset, fine-tuned a GLiClass model, added a pre-inference filter and served a local LLM (LM Studio) behind a FastAPI API.

    decisions

    • Served a local LLM (LM Studio) behind FastAPI rather than calling a hosted API.
    • Python
    • FastAPI
    • GLiClass
    • Transformers
    • LM Studio

  4. Time & mileage app

    Internal mobile app for employees to log hours and mileage, backed by a Go service over a legacy database.

    context

    At Infofil. React Native / Expo front, Go back-end.

    challenge

    Connecting a modern Go back-end to a proprietary HFSQL database over ODBC, with JWT auth, automatic trip geolocation and off-store APK distribution.

    • Go
    • React Native
    • Expo
    • ODBC
    • JWT

about

I'm a developer first: I care more about solving the problem than about which layer of the stack it lives in. Most of what I know about Linux and operations comes from running my own hardware on a Proxmox and LXC homelab.

Away from the keyboard I do amateur astronomy with an electronic telescope, train in powerlifting, and tinker with small electronics projects. The telescope is what got me curious about making software and hardware agree on constrained devices.

contact

Open to backend and systems roles in France and internationally. The fastest way to reach me is email.