Three years shipping production LLM and multimodal systems — fine-tuning, self-hosted deployment, constrained generation, agentic tooling. Depth that runs from model training down through the cloud infrastructure underneath it. Published researcher.
- Work
- Production ML systems, end to end
- Now
- Senior Machine Learning Engineer at Agnisys
- Focus
- Self-hosted infrastructure
- Tools
- Neovim · Arch · Hyprland
- Theme
- Everforest
- Résumé
- Download PDF
AI² — EDA Verification Intelligence
Multimodal document understanding over hardware specifications, grammar-constrained generation into UVM-ready JSON, and a self-hosted Qwen3.6 deployment that replaced the OpenAI API dependency. Cut HLD-to-driver from 30 days to under 6 hours.
Skardyy/mcatmerged upstream
Three interactive-mode bugs fixed in an unfamiliar codebase and merged as PR #83: broken Mermaid rendering, a zoom-triggered panic, and broken navigation.
Parallel ffmpeg transcoding with a semaphore-gated worker pool and GPU probing across five acceleration backends, with software fallback.
- hardikraina079@gmail.com
- GitHub
- github.com/Raina-Hardik
- linkedin.com/in/hardik-raina
- ORCID
- 0009-0007-4281-2989