Bhuwan Prasad Joshi

I am a Computer Engineering graduate from Pokhara University, Nepal. My work focuses on retrieval-augmented generation, information retrieval, low-resource NLP, and evaluation of grounded AI systems.

My current research uses Nepali legal text to study how document structure, retrieval quality, evidence coverage, and answer generation interact in legal question answering. I build end-to-end research pipelines from text normalization and document parsing through BM25, FAISS, hybrid retrieval, grounded generation, and component-level evaluation.

I am seeking a thesis-based MS/MSc research direction focused on knowledge-grounded reasoning, trustworthy RAG, evidence verification, and verification-oriented AI systems.

Research

My research interests are in retrieval-augmented generation, knowledge-grounded reasoning, low-resource NLP, legal AI, and reliable evaluation of AI systems. I am especially interested in treating RAG as a retrieve, reason, generate, and verify process instead of a simple retrieve-then-generate pipeline.

In my current work, I compare passage construction and retrieval strategies for grounded question answering over Nepali statutory text. I study how retrieval failures, missing evidence, unsupported reasoning, and generation errors can be separated through component-wise evaluation.

Research Interests
  • Retrieval-augmented generation and grounded question answering
  • Information retrieval, sparse retrieval, dense retrieval, and hybrid retrieval
  • Low-resource and multilingual NLP, with a focus on Nepali legal text
  • Evidence verification, groundedness evaluation, and error analysis
  • Knowledge-grounded reasoning and trustworthy AI systems

Publications

My research focuses on retrieval-augmented generation, legal question answering, insider threat classification, temporal machine learning, and reliable evaluation.

Selected Publications and Manuscripts
  • Joshi, Bhuwan Prasad and Bhandari, Anshu. (2026). Evaluating Chunking and Retrieval Strategies for Nepali Legal Retrieval-Augmented Generation. The Computer Science Journal (AGH). Submitted.
  • Acharya, Anish, Bhat, Chetan Babu, Joshi, Bhuwan Prasad, and Shahi, Pradeept Jarur. (2026). Insider Threat Classification Using CERT-Derived Features: A Comparison of LSTM, GRU, and TCN. Everest Advances in Science and Technologies (EAST). Submitted.
  • Joshi, Bhuwan Prasad, Bhandari, Anshu, Bhatta, Ishita, and Shahi, Pradeept Jarur. (2025). Legal Divorce Chatbot: A Retrieval-Augmented Generation Based Nepali Legal Question Answering System. IC-AICSE 2026. Submitted.

Projects

I build research and software systems around documents, APIs, AI workflows, and repeatable evaluation.

Nepali Legal Chatbot / LawBot

Built the foundation for a RAG-based Nepali legal chatbot that answers questions from verified legal material. The work includes parsing, chunking, validation, retrieval, and evaluation workflows.

AI Code Reviewer

Developed a full-stack code review tool with separate frontend and backend modules for AI-assisted review workflows and structured review feedback.

Educational Document Assistant

Built a Python and Streamlit assistant that reads uploaded PDFs and generates flashcards, quizzes, and personalized explanations using repeatable extraction, chunking, and validation components.

Stock Market Forecasting Using LSTM and ARIMA

Compared ARIMA and LSTM approaches for stock-price forecasting and built Python preprocessing workflows for sequential financial data.

Experience

Founders NepalMar. 2025-Present

Co-Founder. Develop web applications, API-backed features, and workflow tools using JavaScript, TypeScript, and structured data-validation practices.

Bridge Software SolutionsJan. 2026-Jun. 2026

Technical Product Manager: Backend and AI Systems. Coordinated backend, API, workflow, and AI-system features from requirements through implementation, validation, QA, release, and production support.

株式会社D.BrightSept. 2025-Jan. 2026

AI Automation and System Development. Worked on automation workflows for extracting, summarizing, validating, deduplicating, and routing unstructured business information.

AKARUの特徴May 2025-Aug. 2025

AI Automation and System Development. Developed automation workflows for extracting, summarizing, validating, deduplicating, and routing unstructured business information.

VedyraDec. 2022-Dec. 2023

Founding AI Engineer. Developed knowledge-discovery and learning-support workflows, including prompt behavior, validation patterns, and structured responses from unstructured input.

Outreach & Social Work

I am an active member of Divine Youth Club Nepal, contributing to youth-led service, life-skills activities, and community support work.

Divine Youth Club Nepal
  • Coordinated public speaking workshops and youth development sessions to help participants build confidence, communication skills, and leadership habits.
  • Participated in warm-clothes distribution initiatives supporting vulnerable communities during cold weather.
  • Joined temple-cleaning and community-cleanliness activities to support public spaces, local heritage, and shared responsibility for the environment.
  • Supported youth-service activities connected with social service, life skills, recreation, and community engagement.

Recognition

Recognition: Top 10, Hult Prize National Competition Nepal.
Recognition: Runner-up, Pitch Your Business Idea Season 2.

Personal

My academic interests come from a simple habit: I like understanding why systems work, where they fail, and how they can be made more reliable. I am drawn to research that connects careful problem formulation with practical implementation, especially when the work can support people in learning, decision-making, or access to information.

I enjoy turning unclear problems into structured workflows, experiments, and software systems that can be tested and improved. Across my research, product, and community work, I value clarity, reproducibility, disciplined evaluation, and the patience to inspect failures instead of hiding them.

Long term, I want to contribute to technology that is useful, trustworthy, and grounded in real human needs. I am especially motivated by environments where research, software engineering, mentorship, and service can reinforce each other.