Balaram Neupane
Software Engineer
Results-driven, quick learning & high agency Software Engineer with a varied and complementary skillset.
About Me
I’m a Software Engineer experienced in building scalable backend systems and AI-driven applications, with hands-on work across production ML pipelines, cloud infrastructure, and data-intensive services using Python and modern backend frameworks.
With a strong foundation in computer science and a bias toward ownership, I adapt quickly to new technologies and take systems end-to-end, from design and implementation to deployment and iteration.
Work Experience
Software Engineer
Niural AI
- Architected backend APIs for US payroll management systems, automating year-end tax document generation and processing workflows handling millions of dollars in transactions.
- Developed scalable serverless architectures using AWS Lambda and DynamoDB, managed via CloudFormation and ECS to ensure high availability and security.
- Integrated AWS Cognito, CloudWatch, S3, and ECR to streamline authentication, observability, storage, and deployment pipelines.
- Engineered AI-native APIs to automate complex payroll logic, improving system intelligence and maintainability within a large-scale enterprise codebase.
Machine Learning Engineer
MLExperts
- Led development of a Document Information Extraction system using self-hosted LLMs and custom computer vision pipelines, achieving 90%+ benchmark accuracy.
- Fine-tuned EfficientNet and CNN architectures for real-time screen and document understanding under strict latency constraints.
- Designed and implemented high-throughput backend systems using FastAPI, Django, Redis, and async workflows to support concurrent workloads at scale.
- Productionized the open-source PyPI library unlearn-diff for machine unlearning in diffusion models, focusing on modularity and developer usability.
- Integrated Docker image vulnerability scanning into CI/CD pipelines to proactively eliminate high and critical CVEs in production deployments.
- Owned end-to-end ML feature rollouts including development, evaluation, monitoring, validation, and post-deployment iteration.
- Ensured reliability of AI systems through extensive data-driven evaluation, offline benchmarks, failure analysis, and controlled pre-production testing.
Python Developer Intern
Hyperce LLC
- Developed an AI-driven image processing proof-of-concept for the fashion industry, resulting in an open-source contribution on Hugging Face.
- Prototyped a real-time AR jewelry try-on system using OpenCV and NumPy, focusing on low-latency geometric transformations and landmark detection.
Skills
Languages & Core
AI & Machine Learning
Frameworks & Libraries
Cloud & Infrastructure
Data & Middleware
MLOps & DevOps
Projects
Law Baje
A RAG-based legal assistant for Nepali law using hierarchical semantic search and hybrid retrieval, achieving 80%+ accuracy in legal document grounding while significantly reducing LLM hallucinations.
Unlearn Diff
A PyPI-distributed library for machine unlearning in diffusion models, enabling removal of sensitive concepts from trained weights with minimal computational overhead and strong utility preservation.
Grid Guardian
A predictive maintenance system for electrical transformers using oil-gauge sensor data, built with classical ML pipelines and winner of AI Crusade 2023 (Environment Track).
Holiday Booking Platform
A full-stack production system for booking holidays, flights, and hotels, featuring user workflows, admin dashboards, and backend systems for managing bookings and content.
AI-powered Product Suggestions
An MVP system that recommends commercial products based on raw materials and customization constraints, enabling faster go-to-market decisions for B2B users.
Face Swap & Watermark Removal
A computer vision research project exploring face swapping and watermark removal techniques for fashion and media workflows.
Education & Achievements
Education
Bachelor's of Computer Science and Information Technology
Tribhuwan University, Nepal
High School
SOS Hermann Gmeiner School, Sanothimi, Nepal