Skills
Self-rated, and deliberately short of 100. The gap is where trade-off decisions get made.
Overview across six categories. Pick one to look closer.
Evaluation & Regulated AI
RAGAS, Golden Set Design, Eval-Gated CI/CD, Invariant Enforcement, Decision-Logging Architecture, Observability (Prometheus, Grafana)
Blocks deployment below threshold with an eight-metric golden-set evaluation gate at HealthCrew.ai.
Production AI & Backend
Python, FastAPI, LangChain, MCP, Docker, Kubernetes, Helm, self-hosted vLLM, Server-Sent Events, PostgreSQL, pgvector, Azure, AWS, TogetherAI
Runs Savi's production stack at HealthCrew.ai: self-hosted vLLM streaming over Server-Sent Events, containerized with Docker and Kubernetes.
Retrieval & Graph
Neo4j, Cypher, FAISS, Qdrant, Graph-Constrained Retrieval, Hybrid Retrieval, RAG
Constrains retrieval with graph structure in my MSc dissertation, cutting hallucination rate 94.5% against a naive baseline.
Multimodal Processing
Whisper large-v3, Qwen2.5-VL, GLM-4.6V, SmolDocling, document vision, voice transcription
Reads visa evidence as documents and images in Savi, not just text, grounding decisions in what was actually submitted.
UK Regulatory Frameworks
UK GDPR Article 22, FCA Consumer Duty, NHS DTAC
Enforces the UK GDPR Article 22 boundary as a pre-generation pipeline stage, not a disclaimer.
Core Engineering Discipline
Async Processing, Idempotent Integration Design, Semantic Caching, Hybrid Search
Four years of backend and API engineering at Srinsoft underwrites how I constrain AI systems today, not just prompt them.