Software Engineer and NLP Researcher with 3.5+ years of experience building production-grade microservices and scalable web applications. I combine full-stack engineering with active research in low-resource NLP and machine learning.
- Working as Software Engineer at Astha AI, architecting microservices with FastAPI + Clean Architecture
- Published at EACL 2026 (LoResLM Workshop) on pre-trained language models for low-resource languages
- Contributor to SAFE MCP Framework under the Linux Foundation for AI agent security
- B.E. Computer Engineering from IOE Pulchowk Campus (75.01%)
Languages
Backend & APIs
Frontend
ML & NLP
Cloud & DevOps
Databases
| Paper | Venue | Year |
|---|---|---|
| MaiBERT: A Pre-training Corpus and Language Model for Low-Resourced Maithili Language | EACL 2026 (LoResLM Workshop) | 2026 |
| SimAM-KD: Attention-Enhanced Knowledge Distillation for Efficient Image Classification | TechRxiv Preprint | 2026 |
| The "For You" Page Problem: Explicit Content Exposure and Mental Health Concerns | PsyArXiv Preprint | 2026 |
| Machine Learning Analysis of Tirhuta Lipi | ResearchGate | 2023 |
|
MaiBERT BERT-based language model for low-resource Maithili. 87.02% accuracy on news classification, outperforming NepBERTa and HindiBERT. Published at EACL 2026. |
SimAM-KD Attention-enhanced knowledge distillation framework using SimAM for efficient image classification. Enables lightweight models to retain teacher-level accuracy. |
|
SAFE MCP Framework Contributed 3 security technique specs to the Linux Foundation's SAFE project for securing AI agent tool-use workflows. |
PDF Comparison LLM AI-powered PDF comparison tool using LLM analysis (OpenAI, Anthropic, Gemini) with FastAPI backend. |
|
ML First Principles Complete course from linear algebra to modern LLMs. Learning and teaching ML from the ground up. |
Agricultural Marketplace Full-stack platform for buyers and sellers with ReactJS frontend and Django backend, processing 1,000+ daily transactions. |
SAFE MCP Framework (Linux Foundation) -- Contributed security techniques for AI agent safety:
- SAFE-T1009 -- Authorization Server Mix-up (Critical)
- SAFE-T1308 -- Token Scope Substitution (High)
- SAFE-T1401 -- Line Jumping (High)
- Winner -- Best AI Project at DeltaThon
- Runner Up -- Frontend Web Development Competition (ACES)
- 2nd Runner Up -- Dataverse by DOCSUMO
- Top 5 -- Spirathon (National Level Hackathon by Spiralogics)
- AWS Academy Data Engineering -- Amazon Web Services
- Microsoft Innovative Educator (Azure) -- Microsoft
- Supervised Machine Learning -- Coursera & DeepLearning.AI
Building production systems by day, researching low-resource NLP by night.