ABHYUDAYA POKHREL
Email GitHub LinkedIn Dhulikhel, Kavre, Nepal
Education
Bachelor of Engineering in Computer Engineering
Kathmandu University
Expected 2028
Relevant Coursework
Technical Skills
Programming Languages: Python, C/C++, JavaScript, SQL
Machine Learning & Data: PyTorch, TensorFlow, Scikit-Learn, NumPy, Pandas, Matplotlib
Web & Systems: Flask, React, Flutter, SQLAlchemy
Tools: Git, GitHub, Jupyter Notebooks
Soft Skills: Team Collaboration, Technical Communication, Problem Solving
Experience
Machine Translation Intern
Jul 2026 – Oct 2026
Information and Language Processing Research Lab
  • A three-month internship under the Machine Translation research theme.
  • Contributed to a trilingual Nepali–English–Tamang machine translation project.
  • Continuing research and co-authoring a research paper on this work (in progress).
Projects
Flask • Gemma (multimodal) • RAG • edge-tts
  • Built a pipeline that transcribes photographed Nepali legal documents and splits them into clauses using the document's own numbering.
  • Grounded each clause against the actual text of Nepali law (Labour Act 2017, National Civil Code 2017) via embedding-based retrieval over a cached corpus.
  • Generated plain-language, per-clause explanations with risk flags and citations, streamed to the browser as each clause completes; added on-demand text-to-speech and free-form Q&A over the processed document.
React • FastAPI • Node.js/Express • PostgreSQL • Whisper • MentalBERT • Wav2Vec2
  • Built an end-to-end screening platform that transcribes a user's spoken reflection with Whisper and analyzes both text and audio.
  • Combined MentalBERT text embeddings with Wav2Vec2 acoustic embeddings via late-fusion to classify mental health condition, likely trigger cause, and depression risk level.
  • Implemented the full stack: a FastAPI inference server, a Node/Express + Prisma API and database layer, and a React/Vite frontend.
Python • PyTorch • Streamlit • PyDeck
  • Developed a multimodal AI system for detecting wildfires, illegal logging, and poaching from drone and CCTV feeds.
  • Built an audio classifier capable of detecting chainsaw sounds and gunshots using a dual-model approach.
  • Created an interactive threat map with one-click ranger dispatch using Streamlit and PyDeck.
Other Projects
Open Source Contributions
Leadership & Achievements