Dibij Sharma

I’m an AI/ML engineer and product builder focused on shipping end-to-end intelligent systems across LLMs, computer vision, and real-time applications. I’ve built products where model performance matters, but clarity and usability for real users matter just as much.

In 2024, I founded and built Suisik Boards—an AI-powered LMS for Nepal’s Class 12 NEB board exams. I mapped the publicly available NEB syllabus into a structured question bank for curriculum-aligned practice, then engineered mastery tracking that surfaces each student’s weak areas and recommends targeted practice.

I enjoy taking ideas from prototype to production: for example, I built a document-aware RAG conversational assistant with tool agents for appointment booking and form collection, reducing manual document lookup time by ~60% while keeping end-to-end query responses under 3 seconds. I also developed real-time safety monitoring and drowsiness detection systems that deliver fast alerts (e.g., <1–2 second detection/response behavior) using OpenCV, Gemini’s multimodal API, and MediaPipe-based facial geometry features.

On the ML side, I’ve built models end-to-end as well—like a sentiment classifier trained from scratch with ~84% accuracy and a customer churn approach reaching ~89% AUC. I’m currently exploring RAG architectures, vector databases (Chroma, FAISS), retrieval optimization, and FastAPI/Streamlit Cloud deployments, and I’m seeking remote contractor or internship opportunities in AI/ML, GenAI, and applied research.