OpsPilot — Local-First AI Operations Copilot
A production-shaped local AI system that collects operational evidence before one bounded LLM call, then preserves the investigation, citations, and trace for review.
AI engineer in Paris. I build applied AI systems — LLM orchestration, RAG, agents and evaluation — and I run the production platform they sit on: Linux, Docker, CI/CD, observability. Open to an English-speaking CDI in applied AI, AI software engineering, or the platform work around AI.
Three years of production engineering, now focused on applied AI: I build the LLM systems and I run the platform they sit on.
I build the AI generation, review and publication workflows behind LC Academy's course platform, and OpsPilot — an incident investigator that reads the logs, metrics and runbooks before it answers. I teach the same material to engineering students in Paris.
Since 2024 I own the releases, servers and monitoring behind Linguistic Communication's public and student sites. A bad deploy now rolls back instead of taking the site down.
REST APIs, SQL data models and access permissions, delivered inside a Scrum team.
Master's in Cybersecurity & Cloud Computing at ESILV Paris, on top of a First Class Computing degree from London.
I take work from code to production: releases, servers, monitoring, and the fixes when something breaks. At Linguistic Communication that means keeping the public sites and the student platform online while the team keeps shipping.
Recent builds: a course platform built on OpenMAIC that serves several institutions without mixing their data, AI that answers from real logs and runbooks instead of guessing, and an 82-course catalogue moved into a codebase the team can review. I also teach the same material in Paris engineering schools.
Inbound LAN ports exposed — the private NAS is reached through 1 outbound Cloudflare Tunnel
Local AI models pulled off the LAN and restricted to localhost after closing a network-wide exposure
Causally analysed incidents across TLS, networking, process supervision, and application state
of 500+ participants in an industrial challenge — secure PLM-to-SaaS migration
Architecture, AI, VPS, Docker — every skill below maps to real production work.
Where I've used AI Systems & Guardrails
Portfolio — OpsPilot
Portfolio — Resume Studio
Platform Engineer · Linguistic Communication
Platform and backend engineer in Paris, three years in. This is the part a CV has no room for: what broke, what I’d do differently, and which parts were paid production rather than lab work. Pick a question below, or type your own.

Tashfin
Online · answers in a second or two

Start anywhere: what I shipped, what broke, what I’d do differently. The awkward questions tend to be the interesting ones.
Start with one of these
If it isn’t in my record, I’ll say so — I’d rather that than improvise.
Three walkthrough-backed systems projects spanning local AI orchestration, reproducible cloud data pipelines, and GitOps/Kubernetes delivery.

Local multi-LLM system with consensus, anonymized peer review, chairman synthesis, strict JSON/Zod contracts, and SQLite-backed run history.
Reproducible Spark ML pipeline that runs locally or on GCP Dataproc Serverless with the same config and dashboard-ready outputs.

End-to-end CI/CD and GitOps platform for a multi-tenant SaaS application, featuring GitHub Actions, Docker, Kubernetes, Argo CD, observability with Prometheus/Loki/Grafana, and secure tenant isolation via namespaces and RBAC.
Recent Case Studies
A production-shaped local AI system that collects operational evidence before one bounded LLM call, then preserves the investigation, citations, and trace for review.
Deep Dives
Go deeper into the architecture, delivery choices, and operational tradeoffs behind the builds.
Paris-based and targeting English-speaking CDI roles in applied AI, AI software engineering, and the production platform work around AI.
For employers — how hiring, contracts, and right-to-work routes work for a non-EU candidate in France.