AI/ML Engineer · Kathmandu, Nepal

AI/ML Engineer building intelligent systems that see & reason.

I am an AI/ML Engineer

From deep learning models that read medical scans to self-hosted LLM platforms that reason over enterprise knowledge.

01/About

About me

Computer Engineering graduate from the Institute of Engineering, Tribhuvan University, working where research meets production AI. At NAAMII's MAP-MED Lab I built deep learning pipelines for CT, MRI and X-ray analysis and published work on automated pneumonia detection. Today at Bitskraft I design LLM-powered enterprise systems — hybrid RAG, agentic workflows and secure, self-hosted AI infrastructure.

LLM & retrieval systems

Hybrid RAG, agentic workflows and tool integrations on self-hosted, tenant-aware infrastructure built for privacy and multilingual retrieval.

LangGraph
RAGFlow
LiteLLM
Ollama
FastAPI

Medical imaging & computer vision

Classification, segmentation and localization models for CT, MRI and X-ray data, from annotation and augmentation through evaluation.

PyTorch
TensorFlow
OpenCV
3D Slicer
ITK-SNAP

ML research & evaluation

Reproducible experiments with nested cross-validation, hyperparameter optimization, feature selection and statistical model evaluation.

Scikit-learn
XGBoost
Optuna
LightGBM
Pandas

02/Experience

From research lab to production AI.

IndustryJul 2026 — PresentCurrent
AI/ML Engineer
Bitskraft Pvt. Ltd.
- Developing LLM-powered enterprise AI systems using hybrid Retrieval-Augmented Generation (RAG), agentic workflows and tool-based integrations.
- Designing an AI orchestration layer with FastAPI, LangGraph, LiteLLM and Kong API Gateway, integrating self-hosted LLMs, RAGFlow, embedding models and enterprise APIs.
- Building secure, self-hosted, tenant-aware AI infrastructure with Docker and Ollama, focused on data privacy and multilingual knowledge retrieval.
- LangGraph
- RAG
- FastAPI
- LiteLLM
- Ollama
- Docker
TeachingDec 2025 — Feb 2026
Robotics Instructor
Aakar Association Pvt. Ltd.
- Led robotics sessions on Arduino programming, sensor integration and autonomous systems.
- Guided students building robots with Arduino Uno, IR and ultrasonic sensors, servo and DC motors, and Bluetooth modules.
- Arduino
- Embedded
- Sensors
ResearchMar 2025 — Sep 2025
Research Intern
NAAMIIMAP-MED Lab · Nepal Applied Mathematics and Informatics Institute for Research
- Researched multimodal medical imaging across CT, MRI and X-ray datasets.
- Built deep learning pipelines for disease classification and medical image segmentation.
- Designed preprocessing and augmentation pipelines; annotated and validated datasets in 3D Slicer and ITK-SNAP.
- PyTorch
- Segmentation
- CT / MRI
- 3D Slicer
- ITK-SNAP
Education2020 — 2025
B.E. Computer Engineering
Institute of Engineering, Tribhuvan UniversityKathmandu, Nepal

03/Selected work

Models that matter, measured.

01GUIC 2025

Medical Imaging · Published

Pneumonia Classification & Localization

An end-to-end deep learning system that both detects pneumonia in chest X-rays and localizes the affected region.

ResNet-U-Net acc.: 95%; ResNet-U-Net acc.
MobileNet-U-Net acc.: 93%; MobileNet-U-Net acc.
X-ray images: 20k+; X-ray images

U-Net models with ResNet and MobileNet backbones
Custom IoU loss and dropout regularization for sharper localization
Presented at Gandaki University International Conference (GUIC) 2025

PyTorch
U-Net
ResNet
MobileNet
Computer Vision

02Ongoing

LLM Systems · Bitskraft

Enterprise AI Orchestration Layer

A self-hosted, tenant-aware platform that routes requests across LLMs, retrieval and enterprise tools — keeping sensitive data in-house.

Hybrid retrieval: RAG; Hybrid retrieval
Workflows: Agentic; Workflows
Self-hosted LLMs: Private; Self-hosted LLMs

Hybrid RAG with RAGFlow and embedding models for multilingual retrieval
Agentic workflows orchestrated with LangGraph and LiteLLM
Secured behind Kong API Gateway; deployed with Docker and Ollama

LangGraph
FastAPI
RAGFlow
LiteLLM
Kong
Ollama

03UGC Grant

Medical Imaging · Grant-funded

Early Detection of Knee Osteoarthritis

A five-class classifier that grades knee osteoarthritis severity from radiographs to support earlier diagnosis.

Validation acc.: 87.99%; Validation acc.
Knee X-rays: 8k+; Knee X-rays
Severity grades: 5; Severity grades

EfficientNetB0 enhanced with attention mechanisms
Improved discrimination on moderate and severe OA grades
Funded by the University Grant Commission Mini Research Grant

TensorFlow
EfficientNet
Attention
Transfer Learning

ML Research Tooling

Nested Cross-Validation Framework

A modular, reproducible framework for unbiased evaluation and comparison of ML models on medical data.

Model families: 6; Model families
Feature selectors: 2; Feature selectors
Config-driven: YAML; Config-driven

SVM, Random Forest, XGBoost, LightGBM, ANN and LASSO behind one interface
Nested CV with GridSearch and Optuna hyperparameter optimization
YAML-driven experiments, automated tracking, mRMR and Boruta feature selection

Scikit-learn
XGBoost
Optuna
LightGBM

Published, presented, funded.

Publication
An End-to-End Deep Learning Approach for Automated Classification and Localization of Pneumonia in Chest X-Rays
S. Lamichhane Chhetri, K. Rana Bhat, S. KC, C. Baskota, M. Reezal, K. Raj Koirala, N. Lamichhane · Proceedings of GUIC 2025, pp. 168–169
2025
Conference Talk
Gandaki University International Conference (GUIC)
Presented the pneumonia classification & localization research · Pokhara, Nepal
2025
Research Grant
University Grant Commission — Mini Research Grant
Deep Learning Study for Early Detection of Knee Osteoarthritis from Radiographic Images
2024

05/Skills

The toolkit.

LLMs & Retrieval

Large Language Models
RAG
Hybrid Retrieval
Embedding Models
Agentic Workflows
Multilingual Retrieval
LangGraph
RAGFlow
LiteLLM

ML & Computer Vision

Deep Learning
Image Classification
Image Segmentation
Object Localization
Transfer Learning
Data Augmentation
PyTorch
TensorFlow
OpenCV

Data & Research

Scikit-learn
NumPy
Pandas
NLTK
Feature Selection
Nested Cross-Validation
Hyperparameter Optimization
Statistical Evaluation

Systems & Tooling

Python
SQL
Dart
FastAPI
REST APIs
Docker
Ollama
Kong API Gateway
Git
Linux
3D Slicer
ITK-SNAP

06/Contact

Let's build something intelligent.

Open to collaborations, research opportunities and AI engineering roles. My inbox is always open.

samirlamichhane10@gmail.com