Software & technical architect. Building Engrava, the memory database for AI agents, at my independent AI studio, Sovantica.
Engrava is an open-source, in-process Python library over one SQLite file: a queryable memory graph with bi-temporal valid-time predicates, hybrid (vector + keyword) search, background consolidation from six fixed signals, and a SHA-256-linked journal of thought and edge changes. There is no generative-model call in the core write path. My focus is agent memory that is reproducible and precise about what it does.
Through Sovantica, my independent AI studio, I build open-source memory and retrieval infrastructure for AI agents, with reproducible benchmarking, AI-native engineering (LLM-assisted development plus independent, separate-LLM review), and ADR/RFC governance across my work.
Programming for 25+ years, from C and C++ as a teenager to enterprise financial-services systems. Now building memory and tooling for AI agents.
- AI / LLM: MCP, agentic frameworks, LLM APIs (OpenAI, Anthropic, Gemini), RAG / hybrid retrieval, vector search (pgvector, sqlite-vec), PyTorch, SLM fine-tuning, Hugging Face, Ollama
- Languages: Python, Java, TypeScript / JavaScript, C / C++
- Backend & web: FastAPI, Spring Boot, microservices, REST, HTML / CSS, asyncio, pytest
- Data: PostgreSQL, Oracle, SQLite, Redis
- Platform & DevOps: Linux/Unix, Docker, CI/CD (GitHub Actions, Jenkins), Terraform, AWS, monitoring (Prometheus, Grafana), networking