# Tim Vignon — Portfolio (full content) Generated 2026-09-07 from the site data. Canonical site: https://timvignon.fr Languages available on the site: English, French, Chinese (中文), Korean (한국어). This file is in English. ## Who Systems you can trust. Agents you can watch. Final-year engineering student at UTT. From network intrusion detection and computer vision to an open-source sensor that catches prompt injection on MCP. Final-year engineering student in Networks and Communications at the University of Technology of Troyes (UTT). After a six-month research internship at the LIST3N lab on multi-camera AI surveillance, my research now focuses on runtime security for AI agents: detecting indirect prompt injection on the Model Context Protocol, with mcp-watch as the open-source artifact. I like working across the whole stack: network security and anomaly detection, computer-vision pipelines, cloud and DevOps, and the frontend of a VC intelligence platform built with a three-person team. My studies have taken me across three countries: France, China (Shanghai University, Best ROI & Best Speech awards at the Innovation Bootcamp) and South Korea (UNIST). Beyond that, I build websites for local businesses, tutor international students in French, and pursue interests in volleyball, badminton, entrepreneurship, and astronomy. Key numbers: 7 projects, 3 podium finishes (hackathons and CTF), studies in 3 countries (France, China, South Korea). ## Current focus: Runtime security for AI agents Agents cannot reliably tell instructions from data. A hidden instruction in a document or a tool response can hijack them while they hold legitimate credentials. My master’s research targets indirect prompt injection on the Model Context Protocol, and mcp-watch is its open-source counterpart. - mcp-watch live demo: https://mcp-watch-demo.fly.dev/ ## Projects (7) ### mcp-watch: Runtime Security for AI Agents - Category: AI Agent Security - Year: 2026 - Status: Pre-alpha - Pinned on the globe at: Paris, France - Featured on the home page: yes - Stack: Python, FastAPI, DuckDB, React, Isolation Forest, LLM-as-judge - Links: Live demo https://mcp-watch-demo.fly.dev/ Open-source runtime sensor for MCP-based dev agents: a transparent proxy that baselines tool-call behaviour (Isolation Forest) and scans received messages for indirect prompt injection, correlating both into alerts. Master’s thesis artifact, 150+ tests, live demo. ### Praetor: Physics-Informed Predictive Maintenance - Category: AI Agents × Industry - Year: 2026 - Status: Hackathon prototype - Pinned on the globe at: San Francisco, USA - Featured on the home page: yes - Stack: PyTorch, PINN, Multi-agent, Nemotron Ultra, Crusoe, Time Series Living digital twin predicting industrial machine failures days ahead with no extra hardware: Physics-Informed Neural Networks embed each machine’s equations and infer its state at uninstrumented points. An accountable agent chain (HMAC-signed readings, citable digests, defence/attack/judge agents, LLM jury) vets every alert. Built in 24 hours with a team of five, where I owned the full-stack app and the agentic AI layer. 2nd place at RAISE Summit Hackathon 2026. ### D-Valentin: VC Intelligence Platform - Category: Product Engineering - Year: 2026 - Status: Ongoing - Pinned on the globe at: Singapore, Singapore - Featured on the home page: yes - Stack: Next.js, TypeScript, Tailwind CSS, FastAPI, Gemini, Playwright Evidence-first platform analysing startups and founders through a traceable proof chain (sources → evidence → claims → auditable reports), with pitch-deck analysis, RAG chat and multilingual memos. Frontend owner in a three-developer team. ### Multi-Camera AI Surveillance (LIST3N Research) - Category: Computer Vision Research - Year: 2026 - Pinned on the globe at: Dakar, Senegal - Featured on the home page: no - Stack: Python, YOLO, ByteTrack, Homography, Edge AI Six-month research internship at the LIST3N lab: real-time industrial safety detection with YOLO and ByteTrack on synchronized camera streams, a Bird’s Eye View fusion pipeline projecting detections onto a floor plan via homography, and an edge-versus-cloud benchmark of YOLOv8/11 and RT-DETR. ### DLP Network: Data Exfiltration Detection - Category: Cybersecurity - Year: 2025 - Status: In Progress - Pinned on the globe at: Rio de Janeiro, Brazil - Featured on the home page: no - Stack: Python, Docker, InfluxDB, React, Isolation Forest Open-source network DLP analyzing real-time flow metadata (NetFlow/IPFIX) to detect and block data exfiltration with graduated response. ### Speak My Notes - Category: AI Product - Year: 2026 - Status: Prototype - Pinned on the globe at: Seoul, South Korea - Featured on the home page: no - Stack: Python, FastAPI, Claude API, Expo, EdTech Mobile-first prototype turning photographed language-class notes (Chinese, Korean, French) into immediate speaking practice: vision extraction with the Claude API and a local-first learner memory. ### Freelance Web Development for Local Businesses - Category: Web Development - Year: 2025 - Status: Ongoing - Pinned on the globe at: Dijon, France - Featured on the home page: no - Stack: React, Next.js, Tailwind CSS, UI/UX Bespoke showcase websites for small businesses across Burgundy, with end-to-end ownership from prospecting and UI/UX design to deployment and client management. ## Work experience ### Research & Open Source - AI Agent Security 2026 - Present · mcp-watch (master's thesis project, UTT) · Troyes, France Researching indirect prompt injection in inter-agent and tool communications on the Model Context Protocol: behavioural baselining of tool-call sequences combined with content-level classification of received messages. Building mcp-watch, the open-source runtime sensor behind the research (Python, FastAPI, DuckDB, React), presented at VivaTech 2026. ### Frontend Lead - D-Valentin 2026 - Present · D-Valentin Project (three-developer team) · Remote Owning the web application of an evidence-first VC intelligence platform: Next.js 16, TypeScript, Tailwind CSS, internationalisation in four languages, Playwright end-to-end tests. Working in disjoint ownership lanes with the FastAPI + Gemini backend team, deployed on a VPS with Dokploy. ### Ambassador - Lyon Jul 2026 - Present · Hackathon Hub · Lyon, France · https://hackathonhub.eu/ Part of a 90+ ambassador network connecting Europe's hackathon ecosystems. Recruiting participants and organising events on the Lyon side, with a reach of 700 to 1,000 people through the network. ### Research Intern - Multi-Camera Surveillance Fusion Mar - Aug 2026 · LIST3N Lab, UTT · Troyes, France Developed a multi-camera fusion system for industrial security surveillance: YOLOv8/11 and RT-DETR benchmarking, ByteTrack multi-object tracking, Bird's Eye View fusion via homography, and an edge-versus-cloud deployment study on FPS, VRAM and network latency. ### Freelance Web Developer 2025 - Present · Self-employed · Burgundy, France Designing and building bespoke showcase websites (React, Next.js, Tailwind CSS) for small businesses, with end-to-end ownership from prospecting and UI/UX design to deployment and remote client management. ### French Tutor for International Students 2024 - Present · UTT · Troyes, France Teaching French to international exchange students, helping them integrate into academic and daily life in France. ## Hackathons & CTF ### 2nd place - RAISE Summit Hackathon 2026 Jul 2026 · RAISE Summit / Cerebral Valley, Crusoe track · Paris, France 2nd out of 6,000+ builders from 60 countries, $3,000 prize. Built Praetor in 24 hours with a team of five, owning the full-stack app and the agentic AI layer: PINN-based predictive maintenance with an accountable agent chain and an LLM jury. ### 1st place - TIME0UTT CTF 2026 Jun 2026 · UTT student CTF · Troyes, France Topped the leaderboard with 10,650 points across web, crypto, reverse engineering and forensics challenges. ### 2nd place, Best ROI & Best Speech - Innovation Bootcamp Jun 2024 · UTSEUS - Shanghai University · Shanghai, China Podium finish at the Innovation Bootcamp, with the Best Return on Investment and Best Speech awards. ## Education ### Engineering Student - Networks & Communications (final year) 2022–2027 · UTT · Troyes, France Network and Communications Engineering degree, after the integrated preparatory program (2022–2024). Master's research project on AI agent security; academic projects in cybersecurity, machine learning and cloud. ### Exchange Student - Informatics & AI 2025–2026 · UNIST · Ulsan, South Korea Semester abroad focusing on Machine Learning, Computer Vision and Deep Learning, with an academic project in medical imaging. Courses in English, plus Korean language. ### Preparatory & Innovation Program 2024 · UTSEUS - Shanghai University · Shanghai, China Semester abroad in IT and computer sciences. 2nd place at the Innovation Bootcamp, with the Best Return on Investment and Best Speech awards. ## Skills & tools - AI & ML: Python, PyTorch, TensorFlow, Scikit-learn, Computer Vision, YOLO / RT-DETR, Deep Learning, LLM APIs (Claude, Gemini) - Security & Networks: AI Agent Security, Prompt Injection Detection, Model Context Protocol, Anomaly Detection, Network Security, Wireshark, Cisco CCNA (in progress) - DevOps & Cloud: Docker, Kubernetes, AWS, GCP / Vertex AI, Linux, Bash, Git - Web & Data: React, Next.js, TypeScript, Tailwind CSS, FastAPI, PostgreSQL, DuckDB, UI/UX - Languages: French (native), English (fluent), Chinese (intermediate), Korean (beginner) ## Contact - Email: tim.vignon@gmail.com - GitHub: https://github.com/frisatim - LinkedIn: https://www.linkedin.com/in/tim-vignon-110701260 - CV (PDF): https://timvignon.fr/cv/Tim-Vignon-CV.pdf - Location: Vitteaux / Troyes, France ## About this site Personal portfolio built with React, Vite, Tailwind CSS and react-globe.gl (Three.js). Projects are pinned to cities on an interactive 3D globe; a plane flies between them on the Work page. Visitors can click the globe to leave a marker. Design: sand paper, ink type, carmine accents, day and night themes.