"Intelligence is nothing without accurate retrieval and secure boundaries."
I lead technical vision for AI Engineering, mentoring principal engineers and architecting systems that handle the Hard Trinity of enterprise AI: Scale, Sovereignty, and Security.
After a decade engineering search infrastructure (Solr/Elasticsearch/OpenSearch) at scale, I now focus on what comes next: RAG architectures that survive adversarial pressure, agents that reason over enterprise data, and AI that runs where the data lives.
I stay sharp through CTF competitions—because the best way to build secure AI is to break it first.
| Pillar | What It Means | How I Deliver |
|---|---|---|
| Scale | Systems that handle real enterprise load | RAG platforms indexing 50M+ documents, sub-second retrieval, 10,000+ users |
| Sovereignty | AI that runs without cloud dependency | Air-gapped pipelines, local LLMs, zero-trust architectures |
| Security | Treating LLM safety as adversarial engineering | Red-teaming, prompt injection defense, OWASP LLM Top 10 |
I formalize my open-source research under @ai-search-lab—a "Product Lab" for hardening experimental technology into reusable enterprise patterns.
| Project | Capability | The Pitch |
|---|---|---|
| 🏴 ctf-kit | Automated Red-Teaming | AI-assisted offensive security framework integrating with Claude Code & Copilot for vulnerability detection and exploit synthesis |
| 🧠 adaptive-knowledge-graph | Neuro-Symbolic Learning | Zero-cloud engine fusing Knowledge Graphs with LLM reasoning via Bayesian Knowledge Tracing—consumer hardware only |
| 🧱 agentbricks-experiments | Lakehouse Agents | Architectural primitives for LLMs reasoning directly over Unity Catalog volumes—enterprise data as active knowledge |
| 🎙️ whisper-danger-zone | Sovereign Audio | Air-gapped Whisper + Pyannote pipeline for 100% private speaker-attributed transcription |
Coming next: Extending audio pipeline with Qwen TTS/STT for fully local voice interfaces.
┌─────────────────────────────────────────────────────────────────┐
│ Standard AI Engineering │ Adversarial-First Engineering │
├───────────────────────────────┼─────────────────────────────────┤
│ Build → Deploy → Hope │ Build → Break → Harden → Deploy│
│ "Works on my prompts" │ "Survives hostile inputs" │
│ RAG = embed + retrieve │ RAG + grounding + hallucination│
│ │ detection + guardrails │
│ Trust the model │ Verify, constrain, observe │
└───────────────────────────────┴─────────────────────────────────┘
I maintain active CTF practice (2022→present). The "Danger Zone" repos throughout my profile are deliberate: an Applied Research Sandbox where I stress-test bleeding-edge technology before bringing patterns to enterprise.
Technical Direction
- Define AI architecture strategy across engineering organizations
- Mentor and develop principal engineers in GenAI best practices
- Translate OWASP LLM Top 10 defenses into production guardrails and CI/CD pipelines
Enterprise GenAI Platforms
- Architected centralized RAG ecosystem serving 10,000+ internal users
- Indexed 50M+ documents across engineering and product knowledge bases
- Designed "Federated Model Gateway" abstracting providers (Bedrock, Azure, Local) to prevent vendor lock-in and enable dynamic cost optimization
- Implemented distributed tracing for non-deterministic agent flows
Search at Scale
- Led hybrid search transformation (lexical → semantic → unified) for major e-commerce platforms
- Optimized JVM garbage collection and Lucene segment merging for peak traffic, significantly reducing P99 latency
| Domain | Stack |
|---|---|
| GenAI & LLM | Amazon Bedrock · Azure OpenAI · LangGraph · RAG · GraphRAG · Local LLMs (Ollama/Llama.cpp) · OWASP LLM Top 10 |
| Search & Retrieval | Elasticsearch · OpenSearch · Solr · Lucene · Vector DBs · Hybrid Search |
| Engineering | Python · Java · AWS · Databricks · Unity Catalog · System Architecture |
| Adversarial | CTF · Red-Teaming · Prompt Injection Defense · Adversarial ML |
Recent
- 🎙️ Panel Host · Innovation Day 2025: AI Made Real (Brussels) — Exploring the intersection of vision, creativity, and technology in AI
Conference Talks
- Python Generators for Search Engines · Summer Python Meetup
- Deploying Solr in Multi-Region Environments · Apache Lucene/Solr London
- Effective Molecule Search in Elasticsearch · Cambridge Cheminformatics & Zed Conf
- Browser Fingerprinting & Privacy · Privacy Research
- CTF Competitions · Codeberry Club
Writing (coming soon)
- Building AI that survives adversarial pressure
- GraphRAG vs. naive RAG: when knowledge graphs actually matter
- Local LLM deployment patterns for enterprise privacy requirements
