Where I've deployed

• An internal MCP server giving agents governed access to pre-trade tools, market-data services and team knowledge through one service.
• OAuth 2.1 with PKCE against corporate SSO. Short-lived per-user tokens travel end to end, so every tool call runs under the caller's own entitlements, with a full audit trail.
• Tool annotations drive the harness: read-only tools run freely, state-changing calls stop at a human approval gate, and anything unannotated is denied.
• Typed JSON Schema contracts, recoverable structured errors, response-size limits to protect the context window, OpenTelemetry per tool call, and an eval suite on tool selection and argument accuracy before each release.

MCP, OAuth 2.1, TypeScript, OpenTelemetry, Evals

• Ran requirement sessions on the desk and shipped a React/GraphQL dashboard iteratively against live trader feedback.
• Diagnosed the bottleneck users reported mid-session, rewrote the queries and added a Redis cache.
• Owned it end to end in a regulated environment: market-data integration, release, monitoring and frontline support.

40% faster decisions, $2M a year in cost savings, API responses 60% faster.

React, GraphQL, Redis, Java Spring

• AI social listening across Facebook, Instagram, X, LinkedIn, Reddit, YouTube and TikTok, with scored mentions in one feed in under two minutes.
• A triage pipeline: LLM sentiment, estimated reach, auto-categorisation into the customer's own categories and a reputation score, ranked by reach at risk.
• An MCP server over OAuth 2.0 and scoped API keys, so customers' existing agents drive keywords, mentions and triage through the same service layer as the dashboard.
• Ran the whole loop myself: onboarding calls, four-tier pricing, a no-signup demo, and outreach to the Government of Bihar on listening for flood relief.

Next.js, Node.js, Claude, MCP, OAuth 2.0

• Co-founded and built checkout, door check-in, guestlists and organiser expense settlement.
• Worked the door on event nights: offline-tolerant QR scanning, real-time roster sync across scanners, live triage with door staff, then fed what broke straight back into the roadmap.
• Automated payouts reconciling ticket revenue and shared costs across currencies, aligned to PSD2 surcharging and agent-vs-principal VAT.
• An MCP server so organisers run guestlists, expenses, tasks and payouts by talking to an agent.

Live with Prague venues and labels including Ankali and Harmony Rec.

Next.js, Node.js, MongoDB, Stripe, MCP

• Co-founded Orvio Labs s.r.o. and built the product from the Next.js MVP up.
• Connected to each customer's GA4, Search Console and CRM to attribute AI citations back to pipeline instead of reporting a vanity score.
• A source-graph engine measuring citation frequency by domain across a frozen prompt panel.
• Led enterprise scoping on multi-market programmes and ran the GDPR groundwork for crawling: LIA, Art. 14 transparency, DPIA.

Next.js, GCP, LLMs, GDPR

• Agents handle guest messaging, review replies, issue triage, cleaning and maintenance dispatch, and owner reporting, aiming at about half a traditional manager's fee.
• Event-driven on top of the Beds24 PMS: validated webhooks, tool calls against reservations, calendars, messaging and pricing, and a scheduler that respects the API credit budget.
• Approval gates for refunds, price changes and disputes, with per-property memory of house rules, SOPs and guest history.

Agents, Beds24 API, Webhooks, TypeScript