Hello, I'm
I am pursuing my Master of Science in Computer Science at the University of Illinois Springfield (expected May 2026, GPA 3.8), building on my Bachelor of Engineering in Computer Engineering from Pune University. With over three years of hands-on experience in backend development and distributed systems, I have architected and delivered scalable, high-performance applications serving thousands of users with measurable impact on system reliability and performance.
Work Authorization: Authorized to work in the United States. Eligible for OPT (1 year) + STEM OPT (2 years).
I'm currently pursuing my M.S. in Computer Science at the University of Illinois, building on my B.E. in Computer Engineering from Pune University. With 3+ years of specialized experience in artificial intelligence and machine learning, I've successfully developed and deployed production-grade AI systems that have driven measurable business impact.
At the University of Illinois Springfield, I am developing an AI-powered admissions chatbot using the OpenAI API with custom Drupal integration, implementing features including streaming responses, rate limiting, and persistent session management. This chatbot handles 200+ FAQ queries with 90% accuracy, transforming campus engagement by providing instant, intelligent responses to prospective students.
My expertise spans the entire machine learning lifecycle from data preprocessing and feature engineering to model development, optimization, and production deployment. At Product Dossier Solutions (Kytes), I architected a production RAG system using LangChain, Mistral-7B, and FAISS that serves 10,000+ users across six enterprise clients with 95% query accuracy and sub-50ms retrieval latency. I have built AI inference APIs using gRPC integrated with Spring Boot microservices, handling 110,000+ daily requests with 99.5% uptime. I specialize in TensorFlow, PyTorch, LangChain, and deploying ML models on cloud platforms (AWS, GCP) to serve thousands of concurrent users.
Problem: Real-time American Sign Language translation using computer vision and deep learning.
Solution: Fine-tuned VideoMAE transformer on 239-class ASL dataset, improving accuracy from 62% to 82%
(+32% improvement) through novel Universal Temporal Sub-sampling technique optimized for GPU-constrained training.
Implemented AdamW optimizer and cosine learning rate decay for optimal convergence. Supports multi-sign prediction from
long-form videos with temporal context awareness.
Problem: Note-taking and idea organization is challenging for neurodivergent individuals.
Solution: Developed AI-powered mind map generator using OpenAI Whisper for speech-to-text transcription,
Hugging Face Transformers for key phrase extraction, and Graphviz for automatic visualization. Reduced note-taking time
by 80% while improving information retention through visual organization tailored for ADHD and autism.
Problem: Standard RAG systems perform single-chain retrieval with no source credibility evaluation or confidence scoring, producing unreliable research outputs.
Solution: Built a production-grade multi-agent research system using AutoGen where specialized agents divide the pipeline - one retrieves and searches sources, a second evaluates credibility and assigns confidence scores, and a third synthesizes structured outputs with executive summaries, consensus/disagreement analysis, and numbered citations. Deployed on Amazon EKS with two replicas per service, ALB ingress, and zero-downtime rolling deployments representing a complete MLOps workflow.