Open to research & engineering collaborations

Nitesh
Gautam

I'm an AI/ML engineer and researcher based in Nepal, working at the intersection of production ML systems and applied research. My work spans computer vision, NLP, and multi-agent AI - from identity verification platforms serving banks and financial institutions to agentic workflows and spatiotemporal forecasting research.

Get in touchView work
Python✦PyTorch✦TensorFlow✦HuggingFace✦Transformers✦Computer Vision✦NLP✦LangGraph✦MCP✦RAG Systems✦Multi-Agent Orchestration✦FastAPI✦Docker✦GitHub Actions✦MLOps✦MongoDB✦PostgreSQL✦Python✦PyTorch✦TensorFlow✦HuggingFace✦Transformers✦Computer Vision✦NLP✦LangGraph✦MCP✦RAG Systems✦Multi-Agent Orchestration✦FastAPI✦Docker✦GitHub Actions✦MLOps✦MongoDB✦PostgreSQL✦
01About

The short version

I like problems that sit between a research paper and a deployed product. Most of my days are spent training models that have to survive contact with real users, then tightening them until they hold up under load, edge cases, and the occasional bad input nobody planned for.

Lately that has meant identity verification for banks, agentic workflows that talk to real systems, and uncertainty-aware vision research on the side. I also teach Analysis of Algorithms at Tribhuvan University, write about applied ML, and mentor students finding their footing in the field.

Methods & Tools

  • Python
  • PyTorch
  • TensorFlow
  • HuggingFace
  • Transformers
  • Computer Vision
  • NLP
  • LangGraph
  • MCP
  • RAG Systems
  • Multi-Agent Orchestration
  • FastAPI
  • Docker
  • GitHub Actions
  • MLOps
  • MongoDB
  • PostgreSQL

Education

BE in Computer Engineering

Kathmandu Engineering College, Tribhuvan University

2019 - 2024 · Kathmandu, Nepal

02Experience

Where I've built

AI/ML Engineer I · ThirdFactor AI

Jan 2026 - Present

Nepal

  • Contributing to Prixa's production KYC/identity verification platform serving banks, financial institutions, and ISPs - building OCR (English and Nepali, digital and handwritten), passive liveness detection, and document forgery detection using PyTorch, Ultralytics, HuggingFace, OpenCV, and FastAPI.
  • Driving facial age estimation research: Bayesian ConvNeXt with MC Dropout for calibrated per-sample uncertainty quantification, targeting IEEE Transactions on Biometrics, Behavior, and Identity Science (T-BIOM).
  • Built multi-agent AI workflows using LangGraph, MCP, and Python to automate insurance renewal and quoting for a client - currently in testing ahead of deployment.

Lecturer - Bachelor of Engineering · Tribhuvan University

Jun 2026 - Present

Part-time · Nepal

  • Lecturer for Analysis of Algorithms, a new addition to Tribhuvan University's engineering curriculum.

ML/Backend Engineer · Diyalo Technologies

Jun 2025 - Nov 2025

Nepal

  • Built an agentic ticket reservation system using LangGraph and MCP for BusSewa, a bus ticketing platform serving thousands of daily bookings - enabling natural-language itinerary input with end-to-end route lookup, seat selection, and payment.
  • Designed and developed FastAPI backend services for BusSewa, contributing to core API architecture powering daily ticketing operations at scale.

Co-Founder · Jumbo Eco Vehicles

2019 - 2025

Nepal

  • Co-founded one of Nepal's early electric two-wheeler ventures during the first year of my bachelor's, scaling the business over six years before handing it over to new ownership in 2025.
  • Contributed to Nepal's broader EV transition - Nepal has since become the world's second-largest EV market by new car sales share (73%).
03Work

Selected projects

01

Prixa - KYC & Identity Verification Platform

Production identity verification system serving banks, financial institutions, and ISPs across Nepal. Contributions include OCR for English and Nepali (digital and handwritten documents), passive liveness detection, document forgery detection, and 1:1/1:N face matching.

  • PyTorch
  • Ultralytics
  • HuggingFace
  • OpenCV
  • FastAPI
02

BusSewa - Agentic Ticket Reservation System

End-to-end agentic ticketing system for BusSewa, a platform serving thousands of daily bookings. Accepts natural-language itinerary input and handles route lookup, seat selection, and payment through a LangGraph + MCP agent pipeline.

  • LangGraph
  • MCP
  • Python
  • FastAPI
03

Multi-Horizon Soil Moisture Forecasting - ConvLSTM

Encoder-Decoder ConvLSTM trained on seven ERA5-Land variables to simultaneously forecast soil moisture at 1, 3, 7, and 14-day horizons across Nepal's five physiographic zones. Achieves NSE above 0.97 at 1-day lead and drought detection F1 above 0.88 across all zones.

  • PyTorch
  • ConvLSTM
  • ERA5-Land
  • Geospatial ML
04

Bayesian Age Estimation - ConvNeXt with Uncertainty

Facial age estimation framework producing calibrated per-sample uncertainty alongside every prediction, enabling deployment-time deferral on low-confidence inputs. ConvNeXt-Tiny with MC Dropout under a four-phase progressive curriculum across five benchmarks. MAE 2.76 on Morph-II.

  • PyTorch
  • ConvNeXt
  • Bayesian DL
  • Uncertainty Quantification
04Research

Publications

2025

Beyond Point Estimates: Ordinal Uncertainty-Aware Age Estimation via Bayesian ConvNeXt and Progressive Domain Alignment

Nitesh Gautam

Preprint · Targeting IEEE Transactions on Biometrics, Behavior, and Identity Science (T-BIOM)

2025

Multi-Horizon Spatiotemporal Soil Moisture Forecasting for Drought Detection Across Nepal's Physiographic Zones Using an Encoder-Decoder ConvLSTM Architecture

Nitesh Gautam

Manuscript Finalized · Targeting Journal Submission

05 - Contact

Let’s
build something.

I’m always open to talking about applied ML, research collaboration, or building production systems. The fastest way to reach me is email - I read everything.

Email me
GitHub LinkedIn Medium