Applied Scientist with 2.5+ years specializing in NLP and Vision-Language Models for complex document intelligence and domain-specific reasoning.
I am a Senior Member Data Scientist at Perfios Software Solutions with 2.5+ years of hands-on experience in NLP and Vision-Language Models for document intelligence.
I have proven technical ownership across the full ML lifecycle — translating cutting-edge research into scalable production systems. My open-source work on Kuvera personal finance LLMs has attracted 35,000+ downloads on Hugging Face.
Perfios Software Solutions
Pioneering VLMs (PaliGemma2) with LoRA for financial data; designing agentic reasoning workflows; developing reference-free document legibility algorithms.
Perfios Software Solutions
Reduced inference latency 8s→200ms via distillation; improved table detection by 27.6%; enhanced TSR with semantic row-detection.
IIT Kharagpur
Technical articles on machine learning, reasoning systems, and applied research.
Analyzing dense vs diverse sampling strategies for VLM training on 15k-sample synthetic datasets.
Read More →How I ranked 1st globally in the Reasoning Dataset Creation Challenge using synthetic data.
Read More →An overview of how training regimes evolved from classic approaches to modern reasoning models.
Read More →Lessons learned from production PyTorch development covering best practices and optimization.
Read More →Ranked 1st globally among 150+ teams (Bespoke Labs, HuggingFace, Together.ai). Demonstrated 7B model rivaling 14B–24B models.
Instruction-tuning dataset for Indian financial context. Fine-tuned 8B/14B models with 35,000+ downloads.
Interactive research paper reading assistant with tiered inference (Fast Mode + Deep Analysis) for cost-performance optimization.
Kuvera datasets and models with 35,000+ downloads. Active contributor to Hugging Science (AI-for-Food-Allergies).
AAAI 2026 Workshop (FinForge) and FinNLP @ EMNLP 2025 on behaviour-aware personal finance LLMs.
The Reasoning Course (HuggingFace), Generative AI with LLMs (DeepLearning.AI), ML in Production (Coursera).