AI engineer building reliable autonomous agents and developer infrastructure.
Creator of Grinta — a local-first autonomous coding agent for long-running software-engineering work, recovery, validation, and inspectable execution.
Grinta · Demo · LinkedIn · Email
4 h 33 min autonomous execution · 16,393 recorded events · durable recovery · validation-gated completion
Provider-agnostic terminal agent for long-running repository work. Grinta plans, edits, executes commands, observes failures, and validates outcomes autonomously, with an event ledger, checkpoints, replay, and durable recovery. It integrates local LSP and DAP tooling for code navigation and debugging.
Evidence: v1.0.0 · 4 h 33 min autonomous run · 16,393 recorded events · Raft case study: 39/39 tests · PyPI · Docker
Reusable building blocks for safe, observable, developer-facing AI.
- ShadowGit — Git-backed filesystem checkpoints with safe restore and recovery for coding agents and automation.
- RunBound — cross-platform bounded subprocess execution with timeouts, output limits, process-tree cleanup, and health checks.
- SyntaxGate — code-safety guardrails for reliable editing workflows.
- AgentGuard — focused safety controls for agent execution.
- Fast Indexer — Rust trigram indexer for sub-millisecond substring search in large codebases.
- Public APIs MCP Server — lets AI agents search, probe, and generate integration snippets for 1,500+ public APIs.
- Agent Evaluation Platform — multi-provider LLM-as-a-judge scoring, regression suites, A/B benchmarks, and safety checks.
Contributing upstream to OpenHands / software-agent-sdk:
- #4132 — skip unauthenticated managed-proxy model-info lookup.
- #4134 — retry empty synchronous and asynchronous LLM streams.
- #4136 — bound async state-update publishing for stalled subscribers.
Current status: #4132 is open; #4134 and #4136 are drafts.
Ecosystem: contributing Grinta to MCP, coding-agent, Python, and TUI community indexes; maintaining upstream forks while pull requests are under review.
End-to-end applications where agent workflows become useful, inspectable products.
- BluePrinta — full-stack multi-agent SDLC workspace built with Next.js, TypeScript, Zod, SSE, Prisma, and Gemini.
- SFDR Compliance Validator — document intelligence for financial files: ESG extraction, SFDR checks, data consistency, chart accuracy, and disclosure wording.
- Multi-Agent Research Platform — coordinated retrieval, web research, fact-checking, and synthesis for cited reports.
- Agentic Design Patterns — 21 runnable Python implementations covering planning, reflection, tools, RAG, memory, evaluation, multi-agent coordination, MCP, and human-in-the-loop workflows.
- LangGraph Workflow Orchestrator — examples for conditional routing, parallel execution, approvals, iterative refinement, and checkpoint recovery.
Technical focus: autonomous software engineering · agent reliability · evaluation · developer tools
Stack: Python · TypeScript · Rust · Next.js · FastAPI · Asyncio · Docker
AI Engineering @ ESPRIT · Graduating 2027