class SumanPaudel:
role = "Senior AI Engineer"
location = "Kathmandu, Nepal🇳🇵"
education = "MSc Data Science — Tribhuvan University (GPA: 3.7/4.0)"
experience = "5+ years in ML/AI Engineering"
focus_areas = [
"Production LLM Systems & Agents",
"RAG Pipelines & Document Intelligence",
"Real-Time Voice AI & Speech Processing",
"OCR & Vision-Language Models",
"Self-Hosted ML Infrastructure (vLLM, NVIDIA H100/A100)",
]
fun_fact = "I build things that think, speak, and understand 🧠"
Production RAG + SQL agents for manufacturing analytics using LangGraph & open-source LLMs on vLLM. Achieved 95% accuracy on retrieval tasks with unified reasoning across documents & databases.
Real-time conversational voice systems handling 2000+ daily calls with <900ms latency and 99.9% uptime. Multi-channel SIP trunking, custom TTS integration, and intelligent call routing.
Fine-tuned OCR & Vision-Language models (DeepSeek, Qwen, PaddleOCR) → improved accuracy from 84% → 95% using SFT + DPO. Deployed on 2× NVIDIA H100 SXM via vLLM.
I'm always open to discussing AI/ML, LLM systems, or any cool project ideas.
If you find my work interesting, consider giving a ⭐!