AI Architect · Tech Lead · Python Developer
Building AI systems with a security & audit mindset.

• 🤖 AI Architect / Tech Lead — designing LLM-powered systems and agent tooling (MCP).
• 🏗️ On-prem AI platform, end-to-end — architecture & delivery lead of a closed-contour RAG platform (corporate-document Q&A + OCR + integrations) for a regulated enterprise: import-substitution stack, path to FSTEC-K2 / 152-FZ compliance.
• 🚀 High-load backend & infrastructure — distributed systems, Kubernetes, observability & monitoring, CI/CD.
• 🛡️ Digital audit & AI security — zero-trust APIs, LLM red-teaming, secure authentication.
• 🎓 BSc, Information Security — Computer Systems Security @uniyar.
• 🐍 Python-first. Most of what I build runs on FastAPI + async Python.

Languages & Async

Web & API

AI / LLM

LLM Serving & Gateway

RAG & Retrieval

Document Ingestion / OCR

AI Security & Masking

Vector & Graph Stores

Databases & Storage

Messaging & Tasks

DevOps & Infra

Identity & AuthZ

CI/CD & Code Quality

Observability

Frontend

• On-prem AI platform (private / enterprise) — architecture & delivery lead of a closed-contour RAG assistant over corporate documents (regulations, contracts, scans) + OCR + integrations. Local LLM serving (vLLM + Qwen 3) behind a provider-pluggable LiteLLM gateway with deployment profiles (local / hybrid / closed-contour); hybrid retrieval (Qdrant + OpenSearch + bge-m3 + reranker, RRF) with pre-retrieval ACL; a masking egress-filter (Presidio + Natasha) so only pseudonymized text leaves the perimeter; license-clean BOM (SBOM + SPDX-gate) for boxed resale. Designed for FSTEC-K2 / 152-FZ.
• knowbase — Git-versioned, provenance-grounded knowledge layer extracted from a codebase, served to humans and AI agents via MCP.
• zta-internal-api-prototype — Zero-Trust architecture for internal APIs: mutual TLS, JWT, RFC 8705 certificate-bound tokens; FastAPI reference with a central PDP, attack demos and benchmarks.
• enhanced-jwt-safeguards — Secure authentication on FastAPI: JWT with rate limiting, revocation and refresh.

I work with and contribute to the open-source AI-security ecosystem: PyRIT · garak · NeMo Guardrails · LiteLLM · Giskard · OpenLLMetry