Resume
Forward Deployed Engineer and AI Engineer, Prague, Czech Republic
Full stack and AI engineer with 7 years of experience embedding with users, turning a vague problem into a working system, and keeping it running in production. I build agentic systems end to end: agent harnesses, tool and MCP design, memory and context engineering, evals and guardrails, and the integrations that connect them to the customer's data.
How to look at it
- Scroll to walk in. The trees part and each job steps forward.
- Move your pointer to look around.
- Pick a lens below to light up one kind of work.
- Switch the painting off to read it plain.
Read it for
31 lines. Pick a lens to light up one kind of work.
Download PDFExperience
May 2021 to now
Barclays Investment Group, Senior Full Stack Software Engineer, Markets Pre-Trade Analysis (embedded with the trading desk)
- Building Hive Mind, Barclays' internal MCP server that gives AI agents governed access to pre-trade tooling, internal market-data services and team knowledge, so desks plug their agents into one service instead of building bespoke integrations.
- Auth: OAuth 2.1 with PKCE against corporate SSO, short-lived per-user tokens end to end so every tool call runs under the caller's own entitlements, least-privilege scopes, no shared service credentials, and a full audit trail of every call.
- Harness and tool design: MCP tool annotations drive the harness, so read-only tools run freely while state-changing calls hit human approval gates; typed JSON Schema inputs and outputs, recoverable structured errors, and response-size limits to protect the context window.
- Principles: read-only by default, deny anything unannotated, stateless and idempotent operations, versioned tool contracts, a thin MCP layer over the UI's own service layer, OpenTelemetry per tool call, and an eval suite on tool selection and argument accuracy before each release.
- Primary technical contact for traders, desk heads and business stakeholders: demos, discovery and feedback sessions, requirements, scope and timelines, and release plans for non-technical audiences.
- Embedded with traders to spec and lead a React/GraphQL pre-trade dashboard, shipping iteratively against live feedback: decision speed up 40% and $2M in annual cost savings.
- Fixed a critical market-data bottleneck users reported mid-session: optimised queries and added a Redis caching layer, cutting API response times by 60%.
- Owned the feature end to end in a regulated environment: market-data integration, release, monitoring and frontline support for the desk.
May 2019 to Feb 2022
TEDxKanke, Lead Full Stack Developer (remote)
- Sole technical partner to a non-technical organising team: defined product scope, built and launched a full-stack event platform from scratch, and grew registrations by 250%.
May to Aug 2018
Nationale-Nederlanden, Junior Software Engineer, PeopleSoft HRMS
- Delivered 8 PeopleCode modules for an HR system serving 3,000+ employees across European regional offices, gathering requirements directly from regional HR teams.
- Optimised Oracle SQL queries, making report generation 40% faster and reducing database load at peak.
Jan to Apr 2018
Commerzbank, Junior Software Engineer, core banking applications
- Worked on a Spring Boot banking application serving 100,000+ retail and corporate customers daily.
- Set up Git workflow practices across 15 developers, cutting merge conflicts by 50%.
Side projects
Mar 2026 to now
Social Line, AI social listening platform
- Built and shipped an AI social listening SaaS that tracks keywords across Facebook, Instagram, X, LinkedIn, Reddit, YouTube and TikTok, with pre-scored mentions in one feed in under two minutes.
- Designed the scoring and triage pipeline (LLM sentiment, estimated reach, auto-categorisation, a reputation score) that turns a raw mention stream into a queue ranked by reach at risk.
- Shipped an MCP server over OAuth 2.0 and scoped API keys, so customers' own agents drive the product through the same stateless service layer as the dashboard and REST API.
- Ran the full loop myself: onboarding calls, pricing across four tiers, a no-signup interactive demo, and public-sector outreach to the Government of Bihar on flood and relief monitoring.
Sep 2026 to now
AI-native short-term rental operations, Agentic property management
- Building a rental operation where AI agents run guest messaging, review responses, issue triage, cleaning and maintenance dispatch, and owner reporting, at roughly half a traditional manager's fee.
- Agent layer on top of Beds24 as system of record: event-driven via PMS webhooks with shared-secret validation, tool calls against reservations, calendars, messaging and pricing, and a rate-limit-aware scheduler.
- Human approval gates for refunds, pricing changes and guest disputes, with per-property memory of house rules, SOPs and guest history feeding each agent's context.
- Scoped an owner portal with multi-property accounts, revenue reporting and payouts, plus GDPR and EUR/CZK localisation for a Prague pilot.
Aug 2025 to Oct 2026
Orvio.ai, Generative engine optimisation platform
- Co-founded Orvio Labs s.r.o. and built the product end to end, from a Next.js MVP to a platform that gets businesses cited in ChatGPT, Gemini, Perplexity and Google AI Overviews.
- Deployed into each customer's own stack (GA4, Search Console, CRM connectors) to attribute AI citations back to pipeline.
- Built a source-graph engine measuring citation frequency by domain across a frozen prompt panel.
- Led enterprise scoping and delivery with clients, and ran the GDPR groundwork for the crawling pipeline (LIA, Art. 14 transparency, DPIA, source-URL provenance).
Jan 2025 to now
Partymaker, Nightlife ticketing and event operations
- Co-founded and built ticketing and event operations for promoters running 500 to 3,000 capacity events: checkout, door check-in, guestlists and organiser expense settlement.
- On site on event nights: offline-tolerant QR door scanning, real-time roster sync across scanners, live triage with door staff, and the failures seen at the door fed straight back into the roadmap.
- Automated payouts reconciling ticket revenue and shared costs per event across currencies, with pricing aligned to PSD2 surcharging rules and agent-vs-principal VAT treatment.
- Exposed the platform as an MCP server so organisers run guestlists, expenses, tasks and payouts through an AI agent; live with Prague venues and labels including Ankali and Harmony Rec.
Aug 2023 to Feb 2024
Automated crypto trading platform, Solana trading bot
- Solana trading bot with a 92% success rate on 500+ daily transactions via Web3.js, with OAuth and real-time Firebase sync for 100+ concurrent traders.
Aug 2023 to Sep 2025
SkillNavigator, Professional services marketplace
- Angular marketplace connecting 1,000+ users with geo-location matching and real-time chat on serverless Firebase, cutting infrastructure costs by 85% at 99.9% uptime.
Education
2016 to 2022
Czech Technical University in Prague
B.Tech, Software Engineering
Toolbox
- Agents
- Agent harnesses and loops, multi-agent orchestration, MCP server and tool design, Claude Agent SDK, OpenAI Agents SDK, LangGraph, Vercel AI SDK, human-in-the-loop guardrails
- Memory and retrieval
- Agent memory, context engineering and compaction, RAG, embeddings, pgvector, Pinecone, hybrid search and reranking
- Evals and LLMOps
- Golden datasets, LLM-as-judge, regression suites, Langfuse, LangSmith, prompt caching, model routing
- Product engineering
- TypeScript, React, Next.js, Node.js, Python, Java Spring, GraphQL, PostgreSQL, Redis, Firebase
- Cloud
- AWS, Google Cloud, Azure, Vercel, Docker, Kubernetes, CI/CD, observability
- Payments and compliance
- Stripe, payouts, PSD2 surcharging, VAT treatment, GDPR (LIA, DPIA, Art. 14)
English (fluent), Hindi. Based in Prague with EU work authorisation, and happy to travel to customer sites.