Jyotsana Sharma

About me

I'm an AI/ML Engineer and Researcher currently pursuing my Master's in Computer Engineering at New York University. I specialize in developing cutting-edge AI solutions using Large Language Models, Deep Learning, and Natural Language Processing.

With experience at companies like Coforge and mavQ, I've built production-grade AI systems including clinical trials chatbots, document processing pipelines, and RAG systems that have generated significant business value. Currently, I'm researching audio saliency in egocentric AR/VR at NYU's System & AI Lab while co-founding LetBrief, an AI-powered newsletter platform.

What i'm doing

Machine Learning & AI

Developing advanced ML models using TensorFlow, PyTorch, and scikit-learn for real-world applications.
Large Language Models

Building LLM-powered solutions with RAG pipelines, prompt engineering, and multimodal AI systems.
Computer Vision

Implementing deep learning architectures (CNNs, GANs, U-Net) for image processing and enhancement.
NLP & Data Science

Expert in text classification, document processing, and data analysis using NLTK, Pandas, and NumPy.

Experience

Professional Journey

AR/VR Researcher
NYU System & AI Lab Sep 2025 — Present
Conducting cutting-edge research on audio saliency in egocentric AR/VR environments at New York University's System & AI Lab. Exploring 3D Visual Grounding (3DVG) and EasyCom repositories, leveraging deep learning and generative AI techniques to advance immersive computing experiences.
- Research focus: Audio saliency detection in egocentric AR/VR scenarios
- Technologies: Deep Learning, Generative AI, 3DVG, EasyCom
- Collaboration with NYU's advanced AI research team
Co-Founder
LetBrief Sep 2024 — Present
Co-founded LetBrief, an innovative AI-powered platform that revolutionizes newsletter consumption. Built custom-trained Large Language Models (LLMs) to automatically summarize, categorize, and extract key insights from newsletters, helping users stay informed efficiently.
- Developed custom LLM pipeline for newsletter processing
- Implemented intelligent categorization and summarization algorithms
- Technologies: LLMs, NLP, Machine Learning, Python
Senior AI Engineer
Coforge Sep 2024 — Aug 2025
Led development of enterprise AI solutions, including a clinical trials chatbot using PandasAI and Gemini that interprets JSON/CSV data and provides intelligent answers through graphs, tables, and summaries. Built a sophisticated web-scraping RAG pipeline with LangChain and Vertex AI.
- Built clinical trials chatbot with PandasAI and Gemini
- Developed RAG pipeline using LangChain and Vertex AI
- Boosted client satisfaction by 15%
- Technologies: PandasAI, Gemini, LangChain, Vertex AI, Python
Associate Machine Learning Engineer
mavQ Feb 2022 — Sep 2024
Designed and deployed production-grade NLP and ML models for document processing and classification, achieving up to 92% accuracy. Built advanced solutions including YOLOv5 for checkbox detection, document similarity models, and Text-to-SQL/SOQL pipelines. Engineered deep learning architectures (CNNs, RNNs) for text preprocessing and handling imbalanced datasets.
- Generated $1M revenue through ML model deployment for US state agencies
- Achieved 92% accuracy in document classification models
- Reduced manual processing by 50% with automated solutions
- Enabled 80% structured data extraction with PII masking
- Improved model efficiency by 30% through architecture optimization
- Led SonarQube implementation, improving code maintainability by 25%
- Technologies: Gemini, scikit-learn, TensorFlow, Keras, NLTK, YOLOv5, GPT, LLMs
Full Stack Developer
MTX Sep 2021 — Feb 2022
Architected and developed a comprehensive task management application using Angular.js and Node.js with full CRUD operations. The application directly contributed to a 15% gain in team efficiency and significantly improved overall team morale through better project organization and collaboration.
- Built full-stack task management application
- Improved team efficiency by 15%
- Technologies: Angular.js, Node.js, JavaScript, REST APIs

Projects

DIES - Image Quality Enhancement

Developed DIES, a custom U-Net + Median Filter–based image enhancement model that outperformed GANs and autoencoders in PSNR/SSIM metrics. The system enables high-accuracy license-plate OCR on low-quality images, with research paper currently in review.
- Technologies: Python, U-Net, CNNs, GANs, OpenCV, NumPy
- Metrics: Superior PSNR/SSIM performance vs traditional approaches
- Application: License plate recognition from degraded images
View on GitHub
SnapLink - Cross-Device Gesture Router

Built SnapLink, an innovative cross-device gesture recognition system that synchronizes laptop gestures (scroll, volume control, Do Not Disturb) across multiple devices. Selected as Qualcomm Hackathon finalist, placing in top 30 teams out of 100+ participants.
- Technologies: Python, Computer Vision, OpenCV, Gesture Recognition
- Achievement: Qualcomm Hackathon Finalist (Top 30/100+)
- Features: Real-time gesture detection, cross-device synchronization
View on GitHub
Multi-Class Text Classification

Engineered a sophisticated text classification system using BERTTokenizer and BertForSequenceClassification, categorizing text into five classes. Achieved 80% F1 Score on validation set despite working with imbalanced data, demonstrating robust handling of real-world data challenges.
- Technologies: Python, BERT, Transformers, scikit-learn, TensorFlow, Keras
- Performance: 80% F1 Score on imbalanced dataset
- Techniques: Transfer learning, fine-tuning, class balancing
View on GitHub
AI/Machine Learning

This project is mainly created for my AI/ML fellows, this covers all AI/ML algorithms and libraries fundamentals with their code and explanation along side.
- Key Concepts: Algorithms, Libraries, Fundamentals, Code Examples
View on GitHub