John B.

Full stack and AI engineer. I build AI agents, MCP servers and LLM features that hold up in production: outputs checked by code, evaluations, guardrails, security and cost tracking. Plus the Python and Next.js apps and backends around them. 7 years of experience, master's degree from a top-tier Swiss university. Fixed-price projects, starting with a short paid scoping step: AI agent or chatbot, MVP or SaaS, automation, trading bot. WHAT I DO - AI agents and LLM: agents and MCP servers connected to your CRM, databases and APIs, chatbots and voice agents, RAG on your documents, with outputs checked by code so the model does not invent facts. - Security and reliability: evaluations, guardrails, secrets and access control, hardened Linux servers, and reviews of apps built with Lovable, Bolt or Cursor (Supabase rules, Stripe webhooks, auth, tests). - Web apps and SaaS: Next.js and React front ends, Python (FastAPI) and Node.js backends, PostgreSQL and Supabase, Stripe, user accounts, dashboards, admin panels, deployment on Vercel, Cloudflare and Linux servers. - Automation and data: n8n and Make workflows, integrations with your CRM and APIs, scraping, data pipelines, forecasting and scoring models in Python. - Finance and trading: trading strategies and bots, execution on Interactive Brokers, MetaTrader 4/5, Alpaca, Binance and Kalshi, backtests, risk tools, FIX Protocol. DELIVERED ON UPWORK - AI and LLM: an LLM signal engine running in production (the LLM evaluates daily market data, the code computes every number, email and Telegram alerts, 654 automated tests), AI trading bots, and AI training and evaluation work in French for an AI data company. - Web applications and websites, including dashboards with secure access. - APIs and automation: an automated trading bot on Interactive Brokers, a Kalshi API gateway, Alpaca API integrations with a monthly maintenance retainer, Python backtests turned into live strategies, MT4/MT5 plugins, data extraction from third-party APIs. My work history and reviews are below. OTHER WORK - Pulsar, a trading terminal for MetaTrader 5, sold on the MQL5 Market. - A stock screening and backtesting SaaS with natural language and voice queries, built with a partner. - A content and data platform: Next.js site, FastAPI and PostgreSQL API, scheduled data jobs and five LLM pipelines checked by code. - An AI lighting simulator prototype for a lighting manufacturer, with image generation. - A digital menu and QR ordering app delivered to a restaurant. - A website for a fiber optic company, with AI image processing. - A real estate app that scrapes listing sites and flags underpriced properties. - An AI chess coach prototype whose explanations are checked by code. - Around 50 algorithms and models for funds in Canada, the UK and the US; about $10M in assets advised. HOW I WORK Clear milestones, fixed price or hourly, an update every few days. Every line is reviewed and tested by me before delivery. I tell you upfront what I would not build, and for trading work I never promise returns. Sectors: sales and CRM, e-commerce, clinics and service businesses, real estate, finance. Stack: Python (FastAPI, pandas, NumPy, scikit-learn), TypeScript (Next.js, React, Node.js), PostgreSQL, Supabase, Stripe, MQL4/MQL5, C++, R, MATLAB, Docker, Linux, Vercel, Cloudflare. LLMs: Claude, Claude Code, OpenAI, Gemini, MCP. Languages: English (fluent), French (native). Based in Geneva, Switzerland.