██████╗██╗██████╗      ██████╗ ███████╗
 ██╔════╝██║██╔══██╗    ██╔═══██╗██╔════╝
 ██║     ██║██║  ██║    ██║   ██║█████╗  
 ██║     ██║██║  ██║    ██║   ██║██╔══╝  
 ╚██████╗██║██████╔╝    ╚██████╔╝███████╗
  ╚═════╝╚═╝╚═════╝      ╚═════╝ ╚══════╝

I am a student focused on building predictable, high-performance systems where autonomous AI meets deterministic software engineering. Rather than treating LLMs as stochastic black boxes, my work centers on bounded execution gates, agent regression testing, and token-optimized developer tooling across Python, Go, Rust, and TypeScript.

• 🎓 Education: B.Tech in Computer Science & Engineering (2nd Year)
• 🔬 Open-Source Contributor: Upstream contributions merged in CERN Open Data Portal, Omarchy Linux, HFlow, and CellFence.
• 🛠️ Current Engineering Focus: Deterministic agent evaluation harnesses, bare-metal kernel agent runtimes, and local-first vector RAG pipelines.

• TraceLens: Local-first evaluation and regression testing framework for AI agents (pip install tracelens). Provides inspectable traces, task-level bootstrap confidence intervals, and CI gates with pass@k / pass^k metrics.
• go-dav-os Agent Subsystem: Engineered a "Minimum Working Agent" inside an x86_64 freestanding Go OS kernel. Replaces arbitrary shell evaluation with typed actions, strict allowlists, and execution safety gates.
• AgentLink: Persistent, permissioned, and recoverable execution infrastructure for long-running AI workers with checkpoints and bounded authority.

• RTK (Rust Token Killer): High-performance CLI proxy cutting 60–90% of LLM token consumption by filtering bash outputs before LLM ingestion. Deterministic output ordering.
• zero-shelter: Local-first vulnerability triage tool turning noisy dependency scanner output into an actionable, deterministic fix list without runtime LLM overhead.
• CellFence: CI guardrails enforcing architecture boundaries and static analysis checks on AI-generated pull requests.

• CERN Open Data Portal (cernopendata-portal): Upstream contributor to research data pipelines. Resolved Celery worker process-fork race conditions in invenio-queues, implemented DataCite DOI metadata validations, and streamlined background worker logging.
• Omarchy Linux (basecamp/omarchy): Desktop subsystem integrations, single-flight screen recording concurrency controls, and scoped LocalSend firewall rules for private subnets.

• ReelsIQ (Instagram Reels AI Organizer): Full-stack platform turning thousands of saved video bookmarks into a vector-indexed AI knowledge base. Built with FastAPI, Playwright (session-caching and rate-limit jitter), Whisper audio transcription, PostgreSQL + pgvector, and Next.js 14.

Languages:        Go, Rust, Python, TypeScript, JavaScript, C/C++, Java, SQL, Bash
AI & Retrieval:   LLM Tool Calling, Bounded Agent Runtimes, pgvector, Whisper, RAG, PyTorch Evals
Systems & Cloud:  Linux / POSIX, Docker, Celery, Redis, PostgreSQL, WebSockets, Playwright, CI/CD
Testing & Sec:    Pytest, Deterministic Baseline Testing, Static Code Analysis, AST Parsers

• LinkedIn: linkedin.com/in/siddharth-u-9a07b5385
• Upwork: freelancers/~0108c1285afec08364
• Direct Email: cid066a86@gmail.com