Final year CS student. AI Engineer at Qubicsquare Technologies.
Building production systems that serve 1200+ users daily.
| PR | What I Did | Impact |
|---|---|---|
| #1614 | Fixed race condition in EventLog.append(). File-based locking, patched CVE-2025-68146. 263 additions, 15 commits, 2-round maintainer review |
Eliminated data corruption in concurrent multi-agent setups |
| #1433 | Built early stopping/pruner system for behavior tests — 6 pruner classes, 17 unit tests, integrated across 3 test suites. 15 commits | Cuts LLM cost per test run by terminating on first failure signal |
| #1538 | Improved tool matching on agent resume | Fixed broken agent state after session restore |
| #1539 | Added pause/resume to workspace classes | Full lifecycle control for long-running agent tasks |
| #1564 | DeepSeek v3.2 model support | Expanded model compatibility for the SDK |
#15017 — fix(css): prevent CSS double-bundling
CSS imported in both a page's frontmatter and a component script tag was being bundled twice in production builds. Added content-based deduplication across the style pipeline. 121 additions across 9 files. Reviewed and merged by core team member Princesseuh. Shipped in Astro v5.16.11.
#353 — fix: use safe port range for local cluster postgres instances
Postgres instances were binding to 5432–5434, colliding with local dev installations. Remapped backend to 25432–25434, fixed broken multigateway zone port assignments for zones 2 and 3. 10 commits, 7 files changed, 100% coverage on new lines. Reviewed and merged by @rafael.
aum.sitrai.com — Safe Agent Sandbox Runtime
Autonomous AI agents executing code in Docker sandboxes. 1000+ daily runs, 99.9% uptime. Multi-provider LLM fallback chain (OpenAI → Claude → Ollama). Self-correcting Reflexion Loop bumps task success from ~60% → 92%. Cut costs $400/month, reduced failures 60%.
n0x-three.vercel.app — The Full AI Stack. One Browser Tab. Zero Backend.
First fully in-browser autonomous agent. ReAct reasoning loop via WebGPU inference (MLC/WebLLM), tool use, RAG (transformers.js + voy-search), Python execution (Pyodide/WASM), persistent memory (IndexedDB). 16 quantized models hitting 35–50 tokens/sec on consumer hardware. Zero API keys. Zero data leaves your machine.
groq-coder.vercel.app — AI Code Generation with Groq LPU
Natural language → React/HTML with real-time streaming, multi-stage validation, and instant live preview.
ixchio/agent-sandbox-runtime — Secure Execution Runtime for AI Agents ⭐6
Open-source Docker sandbox with hard resource caps (512MB RAM / 0.5 CPU / 5s timeout), LangGraph state with reflection history and checkpointing. Swarm of 4 specialized agents: Architect, Coder, Critic, Security.
ixchio/AutoPR-AI — Autonomous PR Generation
Benchmarks Llama / Mistral / GPT to auto-generate production-ready PRs with human-like commit messages.
ixchio/go-resilient-commerce — Anti-Fragile Distributed Order Engine ⭐1
Financial-grade order processing in Go. Zero data loss guarantee, built for failure tolerance.
AI/ML: Agent systems · LLM orchestration (GPT / Claude / Llama / DeepSeek) · LangChain · LangGraph · RAG · Vector DBs · Evaluation pipelines · Production LLMOps
Backend: Python · FastAPI · Go · Node.js · PostgreSQL · Redis
Frontend: React · TypeScript · Next.js · WebGPU / WASM
Infrastructure: Docker · Kubernetes / EKS · AWS · CI/CD
Monitoring: Prometheus · Grafana · Weights & Biases
Qubicsquare Technologies — Full Stack AI Engineer (Nov 2024 – Present)
Data pipelines · ML infrastructure · Distributed systems
Graduating May 2026. Available for full-time.
Target: Founding engineer / AI engineer — 0→1, not legacy maintenance.


