Available for hire · Kathmandu, Nepal

NISHAN KHAREL.

I build

Data Scientist and AI Engineer with 2+ year of hands-on experience building machine learning, NLP, and computer vision systems. I turn complex data problems into measurable business outcomes.

View Selected Work Download CV
Portrait of Nishan Kharel
AI Engineer
Data Scientist
0+ Year Experience
0+ AI Projects Shipped
0+ Lectures Delivered
0% Student Pass Rate
Scroll
01 — Selected Work

RECENT
PROJECTS.

Production-grade AI systems I've built — from computer vision pipelines to RAG chatbots and end-to-end ATS platforms.

Bank Note OCR
View Project
Computer Vision Deep Learning OpenCV

Bank Note OCR — Nepali Currency Detection

Real-time Nepali banknote denomination classifier using computer vision and vector similarity techniques, robust under varied lighting and image quality.

View on GitHub
Recruit Nepal ATS
View Project
NLP spaCy NER LLM

Recruit Nepal — AI-Powered ATS

End-to-end ATS pipeline with Tesseract OCR, spaCy NER, and Sentence Transformer embeddings. LLM-based contextual scoring for automated candidate ranking.

Live Demo
Shishu Care Infant Cry Classifier
View Project
CNN-BiLSTM Audio ML

Shishu Care — Infant Cry Classifier

CNN-BiLSTM with attention to classify infant cries (hunger, pain, burping, sleep) from mel spectrograms with near real-time inference.

View on GitHub
RAG Chatbot with pgvector
View Project
RAG FastAPI pgvector Gemini

RAG Chatbot with pgvector + Gemini

Production-ready RAG chatbot using FastAPI + PostgreSQL (pgvector) + Gemini LLM API with Sentence Transformer embeddings for cosine-similarity retrieval. Containerised with Docker-Compose; Streamlit frontend.

View on GitHub
PopUp Vendor Chatbot
View Project
LangChain OpenAI

PopUp — Vendor Conversational Chatbot

Multi-document chatbot with LangChain + OpenAI API, dynamic conversational forms, and session state management for accurate multi-turn vendor workflows.

View Details
Protein Function Prediction
View Project
Deep Learning Bioinformatics

Protein Function Prediction (CAFA-6)

Predicted Gene Ontology labels from amino acid sequences using deep learning, with bioinformatics preprocessing, custom encoding, and class-weighted training.

View on GitHub
AI News Hub
View Project
FastAPI LangGraph Gemini AI Agent

AI News Hub — Live AI Papers & News Agent

A live app that gathers the newest AI research (arXiv, Hugging Face) and news in one place, with an "Ask" agent that pulls real paper and news links on request. FastAPI backend, LangGraph + Gemini agent, deployed on Render.

Live Demo
Nishan Kharel
2nd Place Code for Change Hackathon
02 — About Me

BUILDING
INTELLIGENT
SYSTEMS.

I'm a results-driven AI Engineer and Data Scientist with over a 2 year of hands-on experience deploying intelligent solutions and managing end-to-end AI pipelines.

My expertise spans Large Language Models, Retrieval-Augmented Generation, and data engineering with Docker and modern deep learning frameworks. I build chatbots, implement efficient ETL processes, and deliver data-driven insights that enhance business performance.

Education BSc CSIT · Tribhuvan University
Location Kathmandu, Nepal
Focus Areas LLM · RAG · Computer Vision
Email nkharel57@gmail.com
Download Full CV
03 — Research

RESEARCH &
PUBLICATIONS.

My work is on trustworthy AI for medicine and science — models that stay reliable when the data stops looking like the training set, and that signal their own uncertainty instead of failing silently.

Research Interests

I work on trustworthy AI and foundation models: building systems that stay reliable, calibrated, and adaptable across data modalities — text, images, speech, code, scientific data, and narrow domain corpora. My current focus is medical imaging under distribution shift, biomedical language models, and honest evaluation: bootstrap confidence intervals, ablations, out-of-distribution detection, and reporting negative results as clearly as positive ones.

Model calibration OOD detection Distribution shift Medical imaging Biomedical NLP Retrieval

Publications & Preprints

  1. P1

    Does It Know When It Is Wrong? Classification versus Retrieval for Trustworthy Skin-Lesion Screening on Degraded and Real-World Images

    Kharel, N., & Simkhada, P. (2026)

    Preprint · Zenodo · doi:10.5281/zenodo.21188527

    Read preprint
  2. P2

    Cancer-LLM: A PubMed Central–Derived Instruction Dataset and Fine-Tuned BioBERT Model for Multi-Class Cancer Type Classification

    Kharel, N., & Guragain, A. (2026)

    Preprint · Zenodo · doi:10.5281/zenodo.20935607

    Read preprint
  3. M1

    A Database and Semantic Comparison System for Fire Safety Regulatory Guidance

    Kharel, N., & Simkhada, P.

    Manuscript in preparation, 2026–2027

Research Experience

2026

Trustworthy Skin-Lesion Screening under Degradation and Modality Shift

Independent research · first author · [P1]
  • Compared a fine-tuned EfficientNet-B0 classifier against a DINOv2 embedding plus nearest-neighbour retrieval screener on HAM10000 (10,015 images, lesion-grouped splits), and built a degradation benchmark (blur, JPEG, low light, noise, rotation) on which the classifier's melanoma recall collapses towards zero while retrieval holds 40–50% — a failure aggregate accuracy hides.
  • Validated on PAD-UFES-20 (2,106 real smartphone photographs) with no retraining: retrieval caught 40% of melanomas vs. 11%, and flagged the distribution shift far better (OOD AUROC 0.94 vs. 0.72) — the classifier failed silently.
  • Distance-based isotonic-calibrated confidence reached ECE 0.03 under degradation (classifier: 0.08 → 0.14). Reported bootstrap 95% CIs, a majority-class baseline and ablations, stating non-significant results and one negative finding honestly.
Preprint Code
2026

Cancer-LLM — Biomedical Instruction Dataset and BioBERT Classifier

First author · [P2]
  • Built an open instruction dataset of ~151,800 examples mined from PubMed Central across 9 cancer types and three clinical categories (drugs, symptoms, prevention).
  • Fine-tuned BioBERT for multi-class cancer-type classification, reaching 89.8% accuracy.
  • Selected for the Texas International College mini-research programme, awarded a competitive institutional Mini Research Grant, and presented to external evaluators.
Preprint Code
2026 — Present

Semantic Comparison System for Fire Safety Regulatory Guidance

Co-author · with an M.Sc. Data Science researcher, United Kingdom · [M1]
  • Co-developing clause-level alignment over 2,867 UK fire safety clauses using Sentence-BERT retrieval with calibrated similarity bands, plus a rule-based check that flags changed numeric limits and modal verbs inside otherwise "similar" pairs.
2025 — 2026

Protein Function Prediction — CAFA-6 (Biomedical ML)

Independent project
  • Reframed Gene Ontology annotation as hierarchical multi-label prediction: wrote a GO parser with ancestor propagation over 40,000+ terms and explored ESM-2 protein-language-model embeddings with class-weighted per-ontology classifiers.
Code
2025

Shishu Care — Infant Cry Classification (Healthcare Audio)

Team project
  • Designed a CNN–BiLSTM with attention over mel spectrograms to classify infant cries (hunger, pain, burping, sleep) with near real-time inference.
Code
04 — Professional Journey

EXPERIENCE.

A track record of shipping AI systems across recruitment, document intelligence, and education.

  1. Feb 2026 — Present

    Applied Scientist / Research & Tech Lead Current

    Samni Labs · USA (Remote)
    • Technical lead for HealthCare Suite, a live production care-plan automation system for Washington-state adult family homes: FastAPI + LangGraph on AWS (Bedrock, Transcribe Medical, S3, SQS/SNS) over PostgreSQL/RDS, extracting a ~300-field clinical schema from PDFs and dictated audio with human-in-the-loop review.
    • Hardened it for a healthcare buyer's security review: per-user TOTP MFA with encrypted secrets, least-privilege database roles, an immutable audit trail, automatic redaction of direct identifiers before any model call, and end-to-end HTTPS.
  2. Oct 2025 — Feb 2026

    AI Engineer Internship

    Mantra Ideas Tech · Kathmandu, Nepal
    • Engineered a computer-vision banknote recognition system for denomination classification.
    • Fine-tuned text-to-speech models for Nepali voice synthesis.
  3. Jan 2025 — Sep 2025

    Data Scientist Internship → Full-time

    Mind Risers Technology · Kathmandu, Nepal
    • Built a deployed ATS / resume-matching system (Tesseract OCR, spaCy NER, Sentence-Transformer retrieval with an LLM scoring stage).
    • Built a Nepali fake-news classifier using BERT-based transformers.
  4. Feb 2025 — Feb 2026

    Data Science Consultant Part-Time

    Namaste Network · Kathmandu, Nepal
    • Taught neural networks and transformer architectures (backpropagation, attention).
    • Designed the data-science bootcamp curriculum and teaching materials.
  5. 2024

    Machine Learning Tutor / Teaching Assistant

    Texas International College · Kathmandu, Nepal
    • Delivered 20+ lectures on AI and machine learning.
    • Mentored undergraduates through Python and capstone project work.
05 — Technical Stack

SKILLS &
TOOLSET.

Languages

Python SQL C

ML & AI

Machine Learning Deep Learning NLP Computer Vision Fine-tuning TensorFlow OpenCV Scikit-learn Pandas Jupyter

LLM / RAG / Vector

LLMs RAG pgvector Sentence Transformers OpenAI API Gemini API Prompt Engineering

Backend

FastAPI Flask SQLAlchemy ORM Alembic SSL Configuration

Database

PostgreSQL Vector Database

DevOps & Tools

Docker Docker Hub Git Postman Swagger TablePlus

Certifications

  • Python Programming — Programiz Pro, 2024
  • NumPy for Data Science — Programiz Pro, 2024
  • Structured Query Language (SQL) — Programiz Pro, 2023
06 — What I Offer

SERVICES.

01

AI / ML Engineering

End-to-end machine learning pipelines — data ingestion, model training, fine-tuning, evaluation, and deployment to production APIs.

02

LLM & RAG Applications

Production RAG systems with pgvector, Sentence Transformers, and modern LLM APIs. Chatbots, semantic search, and document intelligence.

03

Data Science & Analytics

Predictive modelling, NLP pipelines, and automated analytics using Python and SQL. Turning raw data into actionable business insights.

04

Backend API Development

Secure, scalable APIs with FastAPI, PostgreSQL, SQLAlchemy, and Docker. Production-grade with proper validation and rate limiting.

07 — Let's Connect

GET IN
TOUCH.

Have a project in mind or want to collaborate? Drop a message — I read every one.

Email nkharel57@gmail.com
Phone +977 986 299 3796
Location Kathmandu, Nepal