Work

End-to-end knowledge and retrieval layer for a production RAG assistant — a deterministic, rule-based ingestion and chunking pipeline (no ML, no LLM), structure-aware indexing for AWS Bedrock, two-stage retrieval with reranking, and a four-pillar evaluation harness.

A spec-driven engine that generates SAP-HANA warehouse objects from YAML and parses existing ones back — bound by a shared intermediate representation and byte-stable roundtrip tests.

A deterministic framework that makes AI-agent software delivery reproducible, auditable, and approval-gated — a no-LLM kernel, single-shot agents, enforced invariants, and a tamper-evident audit log.

A controllable, EU-sovereign environment for agentic coding: LiteLLM tier-routing across on-device and EU-managed providers (any OpenAI-compatible backend swappable in), an adversarial reasoning/execution gate, and a single gateway guardrail.

Metadata-driven batch loader for billion-scale initial loads and million-scale daily deltas in enterprise production.

Established a stable and traceable ETL foundation for department-wide reporting, improving incident diagnosis and reducing operational support effort during reporting cycles.