~ Making machines think, one agent at a time ~
I build Agentic AI systems, RAG pipelines, and computer vision models from Kathmandu, Nepal. Final year Computer Engineering student at Kathmandu University · GPA 3.89/4.0
I'm an AI Engineer from Kathmandu who loves building systems that actually think — not just predict. My work sits at the intersection of LLMs, computer vision, and practical engineering. I've shipped production RAG systems, trained YOLO models for live video streams, built 8-node agentic pipelines, and implemented Transformer architectures from scratch just to understand what's happening under the hood.
When I'm not coding, I'm organizing hackathons, mentoring peers on AI/cloud, or obsessively reading research papers. I believe great AI is built with both rigour and curiosity — and a little bit of chaos.
Multi-Agent Orchestration System
8-node agentic pipeline where each node passes validated state to the next — structured, goal-directed agent architecture. Supports concurrent multi-user sessions with persistent agent state.
View on GitHub ↗
Enterprise RAG with intent-aware query routing — classifies each query (factual / comparative / summary) before selecting a handler, reducing unnecessary LLM calls.
Full supervised + unsupervised ML pipeline on 1,000-patient clinical data — EDA, feature engineering, classification (99.5% accuracy!) and patient risk stratification.
Full Transformer from scratch in PyTorch on 50k+ docs — multi-head attention, positional encoding, SentencePiece. Linear SHAP for token-level explainability maps.
YOLOv8 detecting potholes from live ESP32-CAM feeds, integrated with a web dashboard for municipal authorities. Real-world CV on edge hardware.
Multi-agent robotic system with ESP32s, motor drivers & ultrasonic sensors for search & rescue. Swarm communication via MQTT + camera for target detection & obstacle avoidance.
AR Android app using Vuforia + Google Cloud Anchors to guide users through Panauti's heritage sites with digital storytelling, 3D overlays, and interactive experiences.
Real-time recognition of 25+ ASL gestures with YOLOv5. Won Education Category at KU HackFest 2023 — Nepal's largest hackathon with 500+ participants.
Built Sign AI, a real-time ASL gesture recognition system (YOLOv5), and won Nepal's largest in-person hackathon with 500+ participants.
Competitively selected for an 11-day intensive AI program covering LLMs, agents, geometric deep learning, and AI for healthcare.
Organised AI/ML and cloud workshops for 100+ peers. Built technical communication and documentation habits directly carried into engineering practice.
Coordinated Nepal's largest hackathon across two editions — managing 500+ participants, cross-functional teams, and sponsor relations.
Associate Data Scientist · DataCamp | Supervised ML · Andrew Ng, Coursera | Intro to Statistics · Stanford, Coursera
Currently working on a research paper. Details will be added here once submitted / published. Watch this space — good things take time to bake! ☕
📍 Topic area: AI / Machine Learning
📅 Status: In progress
📖 Venue: TBD
Sharing what I build, learn, and break — from Kathmandu to the internet.
I'm open to full-time roles, freelance projects, and interesting collaborations.
I reply to every email — promise!
Made with ✏️ & ☕ in Kathmandu · aakritipoudel.com.np