Self-extending AI agent that identifies missing capabilities and writes, containerizes, and deploys its own microservices. Features a Skill Factory using FastAPI and Jinja2 with a multi-provider agent loop and Telegram bot for real-time task management.

View project →

High-throughput monitoring layer for ML models inspired by distributed tracing systems like Jaeger. Uses Go and gRPC for low-latency telemetry with a Go-Python bridge for seamless data science integration.

View project →

Backend-agnostic API to run and evaluate code across Daytona, E2B, and Docker through a single interface. Features a normalization layer for cross-sandbox metrics and a Streamlit dashboard for visual benchmarking.

View project →

LLM ensemble system that generates five distinct code variants for a single task and benchmarks them in parallel using Daytona sandboxes against HumanEval datasets. Includes a Variant Marketplace for hot-swapping prompt strategies.

View project →