Building End-to-End ML Pipelines
A practical deep-dive into robust machine learning workflows, from data ingestion to model deployment and monitoring.
Coming SoonLoading Sushan Adhikari's portfolio...
👋Hello, I am
SushanAdhikari — AI/ML Engineer
I build and publish in low-resource NLP, LLM alignment, retrieval-augmented generation, and explainable AI. First author on an ACM ICAIL 2026 paper on Nepali-English legal machine translation, with five total publications across NLP, computer vision, and applied ML. Currently co-founding Nurvexa, an AI medical exam-prep platform deployed in Nepal.
5
Publications
300K+
Users Impacted
2
AI Startups
Scroll Down
About Me
From Research to Production AI Systems
I am a Computer Engineering graduate from Kathmandu University (CGPA 3.90/4.0), with an exchange semester at IIT Palakkad. My work spans low-resource NLP, LLM alignment, retrieval-augmented generation, and explainable AI.
I am first author on an ACM ICAIL 2026 paper on Nepali-English legal machine translation, with five total publications spanning legal NLP, tiny-LLM ethical reasoning, deepfake detection, and physics-informed hydrology — plus ongoing preprints in RAG and explainability.
As co-founder of Nurvexa, I build AI systems that ship: a RAG-based medical exam-prep platform deployed for nursing students in Nepal. I have also built real-time computer-vision inference pipelines at KyraWorks and contributed to Nepal’s national pension platform, which serves 300k+ users.
Education
Kathmandu University (B.Eng.) + IIT Palakkad Exchange
Core Focus
Low-Resource NLP, LLM Alignment, RAG, and XAI
Recognition
ACM ICAIL 2026 (First Author), KUSET Publications, Hult Prize
Stack Used Till Date
Python
PyTorch
Scikit-learn
OpenCV
YOLOv8
NVIDIA DeepStream
Triton Inference
GStreamer
FastAPI
PostgreSQL
Docker
AWS
Python
PyTorch
Scikit-learn
OpenCV
YOLOv8
NVIDIA DeepStream
Triton Inference
GStreamer
FastAPI
PostgreSQL
Docker
AWS
Python
PyTorch
Scikit-learn
OpenCV
YOLOv8
NVIDIA DeepStream
Triton Inference
GStreamer
FastAPI
PostgreSQL
Docker
AWS
Featured Projects
Research Publications
Peer-reviewed publications and preprints in low-resource NLP, LLM alignment, RAG, and explainable AI.
View All on Google ScholarPublished
Directional Asymmetry in Low-Resource Legal Machine Translation: A Nepali-English Case Study
- Built the first public Nepali-English legal parallel corpus (5,024 pairs) and found translation quality is direction-dependent — bidirectional training gained BLEU on Nepali→English but lost it on English→Nepali.
Keywords:Legal NLPMachine TranslationLow-Resource NLP
Published
Enhancing Ethical Reasoning in Tiny LLMs via Fine-Tuning and Multi-Agent Consensus
- Fine-tuned three TinyLlama-1.1B agents — utilitarian, deontological, and virtue-ethics — and combined them via confidence-weighted consensus, validated by a 55-participant human study.
Keywords:Tiny LLMsLLM AlignmentMulti-Agent Systems
Published
Detecting Image Forgeries and Deepfakes: A Comparative Study of CNN and Transformer Models
- Benchmarked CNN and transformer detectors on 140,000+ images spanning classical manipulation and GAN-generated forgeries; InceptionV3 outperformed ViT-Base by 5.51% accuracy.
Keywords:Computer VisionDeepfake DetectionBenchmarking
Published
Physics-Informed Data Augmentation for Sediment Concentration Prediction in Himalayan Rivers
- Benchmarked 10 data-augmentation strategies on 51 years of Himalayan river records; a physics-informed ensemble cut RMSE by 24.3% over the best classical baseline.
Keywords:Physics-Informed MLHydrologyData Augmentation
Preprint
AlgoRAG: Retrieval-Augmented Generation for Theoretical Computer Science Education
- Built a five-stage RAG pipeline over 1,200+ theoretical CS artifacts; found BLEU inadequate for logically-equivalent proofs and proposed a 6-criterion pedagogical quality metric instead.
Keywords:RAGCS EducationComplexity Theory
Preprint
How Much Data Is Enough? Validation Gaps in Manim-Based Mathematical Animation Code Generation
- Built ManimCalc-1493, the largest execution-validated Manim dataset (1,493 scripts across 20 categories), and quantified the data scale needed to close the zero-shot-to-fine-tuned execution gap.
Keywords:Code GenerationLLM EvaluationDataset Scaling
Preprint
Explainability Methods for Hybrid CNN-Transformer Document Layout Analysis
- Compared Grad-CAM, Attention Rollout, and LIME on a hybrid CNN-Transformer document layout model, introducing four new interpretability metrics to quantify boundary alignment and structural consistency.
Keywords:Explainable AIDocument AIInterpretability
My Startups
Nurvexa
RAG-powered nursing exam preparation platform
Growing
Co-founded Nurvexa to help nursing students prepare for licensing exams with adaptive question generation and personalized study plans.
RAGCore Engine
PCLExam Focus
2025Founded
NLPRAGData-Driven Learning
PrepGraduate
AI-powered GRE preparation and score analytics
In Development
Co-founded PrepGraduate to deliver adaptive GRE practice, full-length mock testing, and actionable score insights.
GRETarget Exam
AIAdaptive Practice
2026Founded
LLMsAssessment AnalyticsEdTech
Experience and Education
Jan 2026 - Apr 2026
Junior MLOps Developer
KyraWorks
Promoted from Computer Vision Intern; set up monorepo and CODEOWNERS governance across 10+ team members, SAST/DAST scanning (DefectDojo, CodeRabbit), and 3+ AWS EC2 instances via Terraform.
Oct 2025 - Jan 2026
Computer Vision Intern
KyraWorks
Built real-time inference pipelines with NVIDIA DeepStream and Triton; deployed YOLOv8 across 10+ simultaneous camera streams at ~60 fps aggregate throughput.
Jun 2025 - Oct 2025
Full-Stack Development Intern
National Pension Management System, Nepal
Designed the UI/UX and 50+ database schemas (Spring Boot, PostgreSQL) and integrated liveness detection via Amazon Rekognition for a national platform serving 300k+ users.
Jan 2025 - Present
AI Data Volunteer
Mercuri.world
Sole Nepali member among 150+ global volunteers; designed job taxonomy and pipeline planning for a mental-health employment platform.
Jan 2022 - Feb 2026
B.Eng. Computer Engineering
Kathmandu University, School of Engineering
CGPA 3.90/4.0. Coursework in Computer Vision, Databases, Computer Architecture, and Statistics, alongside first-author AI/ML research.
Jan 2025 - May 2025
Exchange Semester, B.Tech CSE
IIT Palakkad
Completed coursework in Introduction to AI, Linear Algebra, Compiler Design, and Graph Theory and Combinatorics.
My Certifications
Machine Learning Specialization
Andrew Ng — Stanford / Coursera
2026
Andrew Ng's Machine Learning Specialization: supervised and unsupervised learning, neural networks, and applied ML best practices.
Apache Airflow Certifications
Astronomer
2024
Astronomer certifications for Apache Airflow 2 — DAG Authoring and Fundamentals — covering pipeline orchestration and workflow design.
DataCamp Certifications
DataCamp
2025
Tracks in Supervised & Unsupervised Learning, AI Fundamentals, PySpark, and SQL — hands-on ML and data tooling.
Embeddings for NLP
OpenHPI
2025
Embeddings-focused NLP course covering semantic representations and retrieval use cases.
Awards and Achievements
Hult Prize OnCampus Winner
Kathmandu University
Led a computer-vision pipeline for fish-disease classification (Dr.Fish) and represented Nepal at the Hult Prize Regional Summit in Bangkok.
NASA Space Apps People’s Choice Winner
2024
Recognized for Cosmira, an interactive educational web app built for science outreach and engagement.
ICT Awards 2024 Finalist
Rising Star Innovation Category
Selected as finalist for impactful technical innovation and applied AI product execution.
AI Crusade 2023 Winner
Health Track
Won health track with MastiskaTrack, an AI-assisted mental health assessment solution.
Conference and Acceptance Records
USC, NCCI, ICAIL, LEC
Attendance certificates and acceptance proof for ongoing and published research contributions.
Mahatma Gandhi Scholarship
Embassy of India, Nepal
Awarded for academic excellence and strong merit performance during earlier studies.
KU Integration Bee Semi-Finalist
Mathematics Competition
Reached semi-finals in Nepal’s early integration-bee competitions, demonstrating strong mathematical fluency.
Featured In
NASA Space Apps
Hult Prize
Kantipur Media
TechPana
Kathmandu UniversityNASA Space AppsHult PrizeKantipur MediaTechPanaKathmandu UniversityNASA Space AppsHult PrizeKantipur MediaTechPanaKathmandu UniversityMentor Testimonials
My Blog
Machine Learning
NLP
The Art of Fine-Tuning LLMs
Lessons from adapting large language models for task-specific outcomes, grounded in low-resource legal NLP work.
Coming Soon
DevOps
DevOps for ML
How Docker, Terraform, and CI/CD can accelerate reproducibility, deployment velocity, and ML model lifecycle quality.
Coming SoonLet's Connect
Ready to collaborate?
I welcome research collaborations, technical discussions, and interesting engineering problems. If you have something in mind, let us talk.
Location
Dhulikhel, Kavre, Nepal
Phone
Languages
English, Nepali, Hindi