Mik Bry
ENFR

Projects

01 — Selected work

These AI products and software engagements show how I actually work: product definition, architecture, secure delivery, running it in production. Some of it shipped fast because AI did a lot of the typing. All of it went through the same test — does it still hold up once real users, real payments, and real data show up.

Kynto — Hands-on CTO

Dec 2025 – Mar 2026 · Paris · B2B SaaS recruitment with AI

For Kynto, I had three months to go from Figma files to a production recruitment platform. I used four coding agents in parallel, but kept auth, tenant isolation, payments, and the data lifecycle under human review. We shipped in three months — a real-time dashboard, hierarchical job validation, document management, AI-assisted job creation and CV matching, calendar sync, and a data lifecycle built to be GDPR-compliant from the start (export plus 24-month auto-deletion).

Tech details
  • Stack: Next.js 15, React 19, TypeScript, Tailwind, PostgreSQL + Drizzle, Better Auth (multi-tenant), AWS S3, Vercel CI/CD, Vertex AI agents, Stripe, Google/Microsoft Calendar, Lemlist.
  • Scale: 300+ files, 40K+ lines, 17 relational tables, 55 UI components, 40+ Zod-validated API endpoints, 790+ Vitest tests, FR/EN bilingual (13 namespaces).
  • Features: S3 upload with progress tracking, Google/Microsoft Calendar sync.
  • Security: session-based auth, presigned S3 URLs, tenant isolation via Drizzle parameterised queries, OAuth 2.0 for calendars.
  • Workflow: Claude Code (Anthropic CLI) driving 4 parallel AI coding agents, Figma-to-production in 3 months.

Ballpoint — Rust developer, SMB honeypot R&D

Dec 2025 · Paris · Cybersecurity research

Ballpoint needed to catch attackers targeting Windows file-sharing before they reached real data, so I built a honeypot that speaks the SMB protocol itself — zero-copy parsers for NetBIOS and protocol negotiation, not just a log watcher. Getting the protocol details right meant reading through the Samba source to map where the customization surface actually was. It logs and qualifies an attacker's behaviour in real time, with severity-tiered alerts so a security team knows exactly what it's facing.

Tech details
  • Stack: Rust, Docker, SMB1/2/3, NetBIOS; tested with smbclient, nmap, netexec, Wireshark, tcpdump.
  • Scale: full SMB server in 3,000+ LOC.
  • Features: zero-copy binary parsers for NetBIOS and protocol negotiation, severity-tiered alerting (OpenCanary-inspired), HTML + JSON alert reporting, multi-config Docker environments for protocol validation, PCAP analysis.
  • Workflow: audited the Samba C source to map the customisation and extension surface of SMB2/3 — produced a technical write-up on customisation paths (config vs. code modification, GPL constraints).

PlanWork.ai — CTO

Jul 2025 – present · Clermont-Ferrand · AI assistant for renovation and construction

PlanWork.ai helps homeowners plan, budget, and track a renovation, with an AI assistant doing a lot of the legwork. I started with the product itself: who it was for, what the first paid version actually needed, and what could wait. Then I built the SaaS infrastructure, connected the AI, and shipped the first Stripe plan.

Tech details
  • Stack: OpenAI, AWS, Vercel, Next.js, PostgreSQL.

iliadata — Tech lead

Feb 2025 – present · Paris · Cryptography platform

I took iliadata from the specification to a running cryptography platform. The difficult part was protecting authentication, encryption, and storage without making the system impossible to operate. That meant a distributed architecture — a custom C cryptography library compiled to WebAssembly, talking to the backend over WebRTC and WebSocket.

Tech details
  • Stack: React, TypeScript, Tailwind, Shadcn/ui, Node.js, Python/FastAPI, WebRTC, WebSocket, C compiled to WebAssembly via Emscripten, Docker Compose, Scaleway, GitHub Actions.
  • Features: distributed architecture — client plus multi-service backend with secure communications; P2P via WebRTC and WebSocket; custom C cryptography library compiled to WebAssembly and called from a WebWorker.
  • Security: SHA-256 / AES integration, Keycloak auth, homomorphic encryption, secure storage.
  • Workflow: full DevOps — Docker Compose multi-container, automated Scaleway deploy, GitHub Actions CI.

Televista — CTO

Feb 2025 – Nov 2025 · Paris · Digital Signage SaaS

For Televista, I wrote the spec and the response to the public tender for a digital-signage SaaS, then built the MVP itself — video player and admin tools — that had to deliver on what the tender promised.

Tech details
  • Stack: Next.js on Vercel and Azure, video player, Python data-processing tools.
  • Security: Better Auth, GDPR, accessibility and security compliance.
  • Workflow: built with Cursor and Claude Code throughout; generative video/image with Flux, Midjourney and others.

Dylaw — Hands-on CTO

Jan 2025 – Oct 2025 · Paris · Legal-services SaaS with AI

Dylaw had to ship quickly, but it handled lawyers, documents, messages, and payments, so a quick prototype wasn't enough. I built the first version in two to three months, instead of the usual six to eight, and treated auth, data, and payments as production work from the start.

Lawyer profiles, client–lawyer matching, real-time messaging, document management, scheduling, billing and payments, analytics dashboards, and conversational AI assistants automating the legal workflow.

Tech details
  • Stack: marketing site (Next.js, SEO), SaaS app (React), admin panel, REST API (Node.js, PostgreSQL), AI services (Python).
  • Scale: CNB crawler — automated collection across 164 French bars and 78,536+ lawyers, with monthly updates.
  • Workflow: AI-assisted development with Cursor.

Opla — Founder, principal developer

Jan 2024 – present · Clermont-Ferrand · Local AI assistants

I started Opla because I wanted local AI assistants that don't ship your data anywhere. The model runs on your own laptop, full stop.

Tech details
  • Stack: Rust, TypeScript, Next.js, Tauri, WebAssembly, WebGPU.
  • Features: local, secure AI solutions with cross-platform desktop builds; multi-platform libraries and apps built on WebAssembly and WebGPU.

Storyfox — Hands-on CTO

Jun 2021 – Feb 2023 · Paris · Video SaaS

At Storyfox I led the team, hired senior engineers, planned the roadmap, and kept security in the delivery work. My job was to turn business requests into things the team could ship.

Tech details
  • Stack: React.js, React Native, MongoDB, AWS, GCP, video processing, ML.

02 — Open source

This is the same audit work the Trust Report will offer as a paid product — it isn't live yet. Email hello@mikbry.com if you want to know the moment it is.