Demand-intelligence platform and MCP server
- Client
- Recommend
- Period
- Nov 2025 – present
- Role
- Tech Lead / Product Owner
Problem
Turn live demand signals (search, social, editorial, video) into scored objects that marketing teams and AI agents can act on, and make the platform usable from inside the tools people already use.
What I built
- Own the technical direction and product backlog end to end (12+ repositories, 1,300+ commits): ASP.NET Core 9 backend, Python data pipelines, Next.js frontend, Strapi CMS.
- Designed and built the Recommend MCP server (C#/.NET 9) that exposes the platform's research tools to Claude and ChatGPT: OAuth 2.1, scoped API keys, on-behalf-of access policy, audit ledger, OpenAI plugin package.
- Co-built the central auth server (password/Google login, JWT, API keys, OAuth 2.1) and the AI content-chat backend routed through the MCP gateway; hardened the estate with nonce-based CSP, GDPR consent logging and least-privilege data access.
- Led data and AI pilots in Python/SQL: supplier-risk scoring, composed-metric scoring, and a freight-brokerage analysis of 9,000+ orders that surfaced a six-figure yearly margin opportunity from empty-return pairing.
- Set the team's engineering and AI practice: CI quality gates (Biome, Jest, Playwright, CodeRabbit + Claude Code review), Azure-backed release versioning, and a 12-pillar developer-AI standard. Author of the open-source Agentic Trust Layer fact-checker.
Result
Platform intelligence available to AI agents through a standards-based MCP server with OAuth 2.1 and scoped keys. A freight-brokerage analysis of 9,000+ orders surfaced a six-figure yearly margin opportunity. Open-source Agentic Trust Layer: claim verification with calibrated confidence (judge AUC 0.86, ECE 0.056 on the published eval).
Stack
- ASP.NET Core 9
- C#
- Python
- SQL
- Next.js 15
- React 19
- MUI
- Strapi
- Azure
- Docker
- GitHub Actions
- Sentry
- MCP