Yubraj Sigdel

I architect and ship production AI systems at Veel Inc. — AI agents, multilingual moderation, recommendation, and retrieval at scale — and research interpretable AI and NLP for low-resource languages, with a focus on Nepali and open-source tooling.

About

I'm a Senior AI/ML Engineer & Team Lead at Veel Inc., where I architect user-facing AI products end to end with a product-first approach: start from the user and business problem, then design the simplest architecture that ships fast and scales. Most recently I took Veel's own AI agent from a one-page spec to production in about 14 days, on an in-house Model Context Protocol (MCP) server and client that make new capabilities pluggable without touching the agent core.

A recurring thread in my work is trust. I built and deployed an in-house multilingual toxicity moderation model, now live in production, and a multi-signal content moderation system combining policy assessment, brand/product tagging, and authenticity scoring. I care about interpretability: a flagging or pricing decision should trace back to a specific reason, not just a score. Underneath sit distributed, event-driven systems (Kafka, gRPC, Redis, GraphQL) built to stay robust under partial failure.

My research interest is NLP for low-resource languages, particularly Nepali — handwritten-script OCR (ViT + NepBERT), sentence embeddings and semantic search, and Romanized-to-Devanagari transliteration (mT5). I build in the open: the DistantFrames package on PyPI, an FFmpeg MCP server, and models and datasets on Hugging Face. I also lead the AI team and have mentored 8 interns into full-time engineering roles. Open to conversations about AI agents, retrieval, interpretable ML, and low-resource NLP.

Work & Research

Everything, in order

Nothing is hidden — the other track dims.

  1. Apr 2025 — now Role

    Senior AI/ML Engineer & Team Lead

    Veel Inc., On-site · AI Engineer (Apr 2025) → Team Lead (Jul 2025) → Senior (Sep 2026)

    Shipped Veel's own AI agent from spec to production in ~14 days on an in-house MCP server and client. Built and deployed a multilingual toxicity moderation model now live in production, a multi-signal content moderation system, and AI-powered features for Veel's video editor. Architected a low-latency recommendation engine (Kafka, gRPC, Redis, GraphQL), an event-sourced ETL/feature pipeline over 500K+ records, a pricing intelligence system with MLOps retraining, and a semantic video retrieval platform on multimodal embeddings. Lead the AI team and mentored 8 interns.

    • AI Agents
    • MCP
    • Moderation
    • Kafka
    • gRPC
    • Redis
    • MLOps
  2. 2024 Paper

    College Chatbot Using RASA

    KEC Journal of Science and Engineering, 2024

    A conversational chatbot built with RASA for automating student and administrative queries at Kathmandu Engineering College.

    • RASA
    • NLP
  3. May 2024 — Apr 2025 Role

    Junior AI Engineer / AI Trainee

    Vertex Special Technology (P.) Ltd., On-site

    Built GenAI-powered, RAG-based Gherkin test generation pipelines and automated DOM/XPath/Selenium workflows, cutting test execution time by ~40%. Contributed image embeddings and CV pipelines to TelePlantDoctor.

    • RAG
    • Selenium
    • Computer Vision
  4. Jul 2024 — Sep 2024 Volunteer

    Contributor, AI for Women Farmers

    FruitPunch AI, Remote

    Annotated Nepali documents for OCR, NLP, and NER systems, and collaborated with OCR teams to strengthen Nepali text identification.

  5. 2025 Project

    DistantFrames

    Open-source Python package · PyPI

    Frame extraction that samples on visual change rather than fixed intervals, using HSV histogram correlation against the last kept frame, with an optional open-eyes filter. Built for computer vision preprocessing, dataset generation, and moderation pipelines; published on PyPI (pip install distant-frames).

    • Python
    • Computer Vision
    • OpenCV
    • PyPI
  6. 2025 Project

    FFmpeg MCP

    Model Context Protocol server

    An MCP server exposing 12 composable, validated FFmpeg tools — clipping, cropping, scaling, overlays, frame and audio extraction, GIF generation, and transition-based concatenation — so language models can run media workflows from natural-language instructions. Each tool returns an output path or a structured JSON error the model can recover from. Runs in Claude Desktop, Cursor, and Cline.

    • MCP
    • FFmpeg
    • FastMCP
    • Pydantic
  7. 2024 Project

    Handwritten Document Conversion

    Open-source · GitHub

    A ViT + NepBERT two-stage architecture for handwritten Devanagari digitization, with YOLOv8 word-level detection, achieving a 9.08% Character Error Rate.

    • ViT
    • NepBERT
    • YOLOv8
    • RoBERTa
  8. 2024 Project

    Roman English to Nepali Transliteration

    mT5 · Hugging Face

    An end-to-end Romanized Nepali → Devanagari transliteration platform, with fine-tuned model and dataset published on Hugging Face.

    • mT5
    • Hugging Face
  9. Apr 2023 — Jul 2023 Internship

    AI/ML Intern

    FuseMachines Nepal, Hybrid

    Built a context-based Nepali News Recommendation System from scratch — scraping, embedding pipelines, and deployment — and trained a Nepali Sentence Transformer using translated STS datasets over NepaliBERT baseline embeddings.

    • Sentence Transformers
    • NepaliBERT

Publications

Peer-reviewed & preprints

  1. Yubraj Sigdel, Sujan Shrestha, Nabin Neupane. College Chatbot Using RASA. KEC Journal of Science and Engineering, vol. 8, no. 1, 2024. DOI Code

Full list on the academic portfolio →

Education

Degrees & schooling

  1. Bachelor's in Computer Engineering Kathmandu Engineering College, 2018–2023.

    82.41%
    • Minor Project: a computer vision system for face detection and recognition applied to criminal identification.
    • Major project: a Network Intrusion Detection System benchmarking multiple ML algorithms.

    Coursework Artificial Intelligence (AI), Mathematics, Applied Mathematics, Probability & Statistics, Numerical Methods, Big Data Technologies, Data Mining, Database Management System (DBMS), Operating Systems (OS), Data Structures & Algorithms (DSA), Theory of Computation (TOC), Object-Oriented Programming (OOP)

  2. High School Gorkha United Public School, 2016–2018.

More on the academic portfolio →

Also

Skills

  • System design AI System Architecture, Distributed & Event-Driven Systems, Microservices, Hexagonal Architecture, Product-First AI Development, Technical Leadership
  • AI / ML LLMs, Generative AI, AI Agents, Agentic Workflows, RAG, NLP, Computer Vision, Recommendation Systems, Semantic Search, Predictive Modeling, Fine-Tuning, AI Content Moderation, Interpretable AI, Low-Resource Language Modeling
  • Retrieval & search Embedding Pipelines, Semantic & Hybrid Search, Graph-Based Retrieval, Vector Indexing
  • Frameworks PyTorch, Hugging Face Transformers, LangChain, LlamaIndex, FastAPI, Sentence Transformers, Scikit-learn
  • Backend & systems Kafka, Redis, GraphQL, gRPC, REST APIs, MCP (Server & Client), FFmpeg
  • Data & databases PostgreSQL, Milvus, ChromaDB, pgvector, Pandas, NumPy, Selenium, BeautifulSoup, Feature Engineering
  • Languages & DevOps Python, SQL, Bash, Docker, AWS, Git/GitHub, CI/CD, Production AI Deployment