Botcrypto.io — No-code algorithmic trading SaaS platform

Problem

During the crypto bull run, there was no simple, affordable way to create automated trading strategies. The need was clear: let retail traders buy and sell 24/7, following preset rules, without any programming skills.

Solution

We designed and built Botcrypto.io — an innovative SaaS that was, for a time, the only platform letting users visually craft their own strategies and automate their execution on Kraken and Binance. A no-code, ergonomic product built to democratise algorithmic trading.

Project duration

4 years — from inception and architecture design to continuous development.

Impact & Key figures

ImmoDuo.fr — Real estate search SaaS platform

Problem

Searching for a home as a couple is a real organisational challenge. With listings scattered across dozens of sites, it quickly becomes hard to centralise information and track the status of visits and appointments.

Solution

ImmoDuo lets couples centralise and record every listing they come across during their search. It's the ideal tool to structure a property project and track progress in real time — with an AI assistant on hand to help along the way.

Project duration

1 year — from inception and architecture design to continuous development.

Development workflow

A Claude Code agent pipeline handles most of the day-to-day: RICE triage on incoming GitLab issues, tech planning posted back as an issue comment, then implementation in an isolated worktree up to the merge request. I stay in the loop to review and approve.

Impact & Key figures

Screenshots

NoTypo — French grammar correction that teaches you the rule

Problem

Most grammar tools correct in isolation: you paste text, you get corrections, you move on — and make the same mistakes next week. French agreement, conjugation and accents are unforgiving, and generic spell-checkers miss the subtler grammatical mistakes entirely.

Solution

Paste French text and Mistral AI returns an annotated list of mistakes — each with the original word, the suggested correction, the rule that was broken, and its category. Every correction is saved and categorised, so a personal error profile takes shape: how many mistakes per category, which specific rules you trip on most often, and your history over time. You learn whether it's accords du participe passé, conjugation, or plain typos — instead of a vague sense that your French could be better.

The learning loop

A spaced-repetition quiz turns that error history into practice material. It picks rules from your own weak spots and drills them at increasing intervals — the classic method, applied to your actual mistakes rather than a generic syllabus.

Keeping the model honest

Mistral isn't infallible. A wrong suggestion can be flagged directly in the correction card, and those reports accumulate as the primary dataset for prompt tuning: high-frequency flags reveal systematic errors — a wrong rule applied, a hallucinated correction — and guide the next iteration.

Stack

Rails 8.1 API-only on Ruby 3.2, PostgreSQL, Devise + JWT for auth. The React 19 + TypeScript + Vite SPA compiles into Rails' public/ and is served as static files, so production runs a single Rails process — no separate Node server. Deployed with Kamal 2.

Project duration

2 days — from blank repo to a deployed, publicly usable app.

Getting started

One command with your own key: git clone https://gitlab.com/cdlr75/notypo && cd notypo && MISTRAL_API_KEY=your_key docker compose up. The database is created and migrated on first run.

Demo

CV.ai — AI-powered resumes tailored to each job offer

Problem

Most job seekers maintain one generic CV and tweak it manually for each application. Existing open-source alternatives either lack AI generation (Reactive Resume, JobSync) or are too heavy to self-host quickly (ResumeLM). None combined a master profile with per-offer AI generation.

Solution

Fill in your master profile once. Paste a job offer. The app generates a tailored resume and cover letter optimized for that specific role. Review and edit in-app, export as PDF, and track every application through Wishlist → Applied → Interview → Offer / Rejected.

Project duration

1 day — from blank repo to a fully working AI resume generator.

Getting started

Single Docker command: docker build -t cv.ai . && docker run -p 3000:3000 -v "$(pwd)/dev.db:/app/dev.db" cv.ai. No external services required beyond an AI API key.

Demo

Nosheet — A focused project tracker for elected officials

Problem

Mayors and local officials juggle many small projects in parallel — road repairs, school renovations, grant applications, meeting follow-ups. Off-the-shelf tools are either too heavy (Jira, Asana, Monday) or too generic (a spreadsheet). Basecamp is great too, but its free tier allows a single project and the data is hosted abroad.

Solution

A purpose-built, local-first project & task tracker — as simple as a spreadsheet, but actually shaped around projects, tasks, and notes. Dashboard, subjects to group related projects, markdown notes with live preview, command palette, quick-add, undo, optional password protection, and full English/French interface. Data lives in a single CSV file the user owns.

Project duration

2 days — from blank repo to a deployed, daily-used app.

Deployment

Self-hosted on my homelab with Kamal. In production for my mother, mayor of Villentrois-Faverolles-en-Berry (36600).

Limitations (honest)

Single user · manual CSV backup (via Settings → Backup) · scales comfortably to ~300 projects on a single CSV file. Designed for one person managing many small projects, not for teams.

Demo

Screenshots