AI Engineer | Data Scientist (ML focus) | LLMs, RAG & Multimodal AI | Python • AWS • MLOps | MS Applied Data Science @SJSU
I'm a Machine Learning Engineer with 3+ years of experience building scalable ML systems, real-time inference pipelines, and Generative AI applications across NLP and computer vision domains. Experienced developing low-latency production workflows using AWS, Airflow, Docker, and distributed data pipelines. Specialized in LLM orchestration, multimodal AI, RAG architectures, and production ML deployment for large-scale applications.
🔭 Current Focus: Building applications with Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), and deploying predictive models.
📚 Key Skills: Python (PyTorch, TensorFlow), Hugging Face Transformers, NLP (BERT), Computer Vision (CNNs), and AWS (SageMaker, Bedrock, Glue, S3).
🎯 Passionate About: Translating complex data into actionable insights and building reliable, scalable AI systems.
🤝 Let’s connect: LinkedIn
📰 IEEE CAI 2025 Published Paper
📌 Built a multimodal AI system using YOLOv8, BLIP-3, and LLaMA3 for personalized advertising based on images, audio, and transcripts. Integrated vector search and GenAI-based personalization.
🤖 Fine-tuned Microsoft's Phi-3-mini LLM via QLoRA (8-bit quantization & LoRA adapters) on attention projections for robust persona alignment. 🔍 Devised a RAG pipeline processing 10GB+ data using FAISS vector search and BERT embeddings, increasing response faithfulness and context precision by 35%.
💹 Developed an unsupervised anomaly detection model using K-means clustering and Locality-Sensitive Hashing (LSH) on streaming data, achieving 95% accuracy. 🔒 Applied differential privacy (K-Anonymity) and Bloom filters in a Spark pipeline to anonymize 5000+ transaction records.
🧠 Architected a multimodal computer vision ensemble by fine-tuning EfficientNet and integrating OpenAI CLIP embeddings. 📊 Achieved a balanced 41% F1-score across 6 distinct inventory classes and streamlined an image processing pipeline for 500,000+ warehouse images
⚡ Optimized EV charging and home energy use using ML models. Integrated weather, solar, and consumption data into a predictive scheduling system.
-
📊 AWS-based Sentiment Analysis Pipeline (3-min Demo)
Brief walkthrough of building a scalable NLP pipeline using AWS Glue, S3, and Redshift with a Streamlit dashboard. -
🎙️ Interview with Industry Expert – Sunny Malik, Sr. Solution Architect @ Snowflake
Conducted as part of a group assignment under Big Data Course, this interview dives into modern big data technologies and trends shaping the data landscape.
📍 San Jose, CA, USA 🔗 LinkedIn