Agnostic, composable AI engines
In production
Built at EsePlus / Alilo · Head of AI Platform Engineering
Two AI engines I designed and own end-to-end, exposed as agnostic, composable APIs for the product team to consume into experiences. One is an async generative-AI backend that turns uploaded documents into corporate microlearning — a multi-stage pipeline coordinating document analysis, content generation, and activity design, with backpressure handling and streaming feedback. The other is a multi-tenant RAG conversational assistant — hybrid retrieval and reranking, modern embedding strategies, an agent runtime with multimodal tool calling, and a defense-in-depth safety stack validated by automated adversarial regression.
Highlights: 95% → 99.5% availability uplift · 20%+ cloud cost reduction · ~90% token savings on validation retries · sub-3s p50 latency · zero jailbreaks in adversarial testing.
Clarytia — verifiable internal-knowledge AI
Private beta
My own product · designed, built and audited solo
A multi-tenant RAG platform that turns a company's own documents into an assistant that answers with the source, the page and a confidence tier — and says "I don't know" rather than improvising. Database-per-tenant isolation, so "delete everything about this customer" is a DROP, not a query I have to pray covers every table. Five ranking lanes fused by Reciprocal Rank Fusion in a single SQL query — no external search service, no extra network hop. Access control is enforced inside the retrieval path, not as a filter afterwards. And it does something most knowledge tools don't: when it genuinely can't answer, that becomes a clustered, named knowledge gap — telling you what your documentation is missing, ranked by how many distinct people hit it.
Highlights: 2,814 automated tests at >90% coverage · 37 architecture decision records · 105 API endpoints · zero leaks across a generative red team gated at every deploy · built in 5 months, single author.
Self-service platform for AI SRE agents
In production
Enterprise engagement · specifics under NDA
A platform that provisions a team-scoped incident-response agent — its integrations, its permissions and its behaviour — from a single block of configuration. Onboarding a team is a merge request, not a project. The part I'd argue hardest for: the agent's playbooks and standing instructions ship as versioned packages pinned per agent, so changing how an agent behaves is a reviewed diff with an approver — not a paste into a console text box nobody can audit or roll back. Every integration is read-only by construction: the agent recommends, humans act.
The bet underneath it: what you control is not the model — it's the model's context and its instructions. The platform operates zero models of its own, on purpose.
Shared CI/CD delivery platform
In production
Enterprise engagement · specifics under NDA
Not a pipeline — a delivery platform. A library of reusable CI/CD components (install, lint, test, build, deploy) that any repository in the organization consumes in three lines of config, with three consumer repositories running on it in production. Branch-routed deployment to per-team sandboxes, DORA metrics instrumented end to end, and a performance refactor that came out of measurement rather than instinct: the quality gate moved downstream, because the build never read lint's output — lint blocks publishing, not compiling.
Highlights: −52% pipeline wall clock · −54% runner time · −62% jobs per pipeline — measured p50 over a 14-day window, not a lucky run.