Storyline AI 

Verusen AI is a supply-chain intelligence company leveraging artificial intelligence to unify, optimize, and automate material management across global enterprises. During this internship, I contributed to the development of scalable AI systems for material classification, supply optimization, and ontology-driven knowledge representation.

Key Responsibilities & Contributions:

• Developed and deployed Verusen Material AI (VMAI) an enterprise-grade inventory insights platform enabling data-driven material management decisions.

• Fine-tuned transformer architectures for automated material classification into 27 proprietary categories, combining rule-based logic to achieve 86% accuracy and reduce external API costs by 67%.

• Engineered a hierarchical UNSPSC categorization model (57 segments → 566 families → 7,997 classes → commodities) by fine-tuning 310M-parameter models, integrating LangGraph APIs and human-in-the-loop feedback for adaptive performance.

• Designed scalable NLP preprocessing pipelines on Azure Databricks + Apache Spark, integrated with Jenkins/GitLab CI/CD, and optimized distributed GPU/CPU clusters, reducing large-scale training and deployment time by 16%.

Storyline AI 

Storyline AI is an agentic behavioral intelligence system designed to assist patients and support healthcare professionals. It interacts naturally with patients, capturing subtle behavioral cues, such as facial expressions, vocal frequency changes, and linguistic confidence, to assess emotional and physical well-being. The system integrates real-time multimodal sensing and emergency alerting, ensuring timely medical responses and improved patient assurance.

Role: Focused on Image Processing Model Optimizing
Technologies Used: OpenCV, PyTesseract

Key Responsibilities & Contributions:

• Developed and optimized Retrieval-Augmented Generation (RAG) pipelines to enhance AI-driven multi-agent performance.

• Built Agentic AI systems using LangChain, LangGraph, OpenAI, and OpenSearch, improving contextual understanding and patient engagement.

• Leveraged computer vision to extract over 30,000 behavioral features, integrating them with DALL·E-based representations for real-time patient monitoring.

• Implemented range and term filters in OpenSearch, reducing query latency by 15% and improving system responsiveness.

• Conducted research on Generative AI evaluation using RAGAS, refining chatbot accuracy, coherence, and reliability in healthcare contexts.

SeamOCR 

SeamOCR is an ongoing project aimed at streamlining digital invoice processing through advanced image and OCR technologies. The project involves two main components: an image processing engine and an OCR engine, working together to convert invoice images into machine-readable text

Role: Focused on Image Processing Model Optimizing
Technologies Used: OpenCV, PyTesseract

Responsibilities:

• Fine-tuned the image processing model to enhance OCR (Optical Character Recognition) results significantly.

• Standardized parameter ratios, achieving a 65% improvement in OCR accuracy, thereby enhancing the reliability and efficiency of invoice processing systems.

FDE Scraper 

The FDE Scraper project focuses on the continuous collection, cleaning, and storage of financial data from 600 different websites, covering 5000 financial funds. This data is essential for providing up-to-date financial insights to clients.

 
Role: Data Cleaning and Management
Technologies Used: SQL, Various NLP Technologies
Project Description:

Responsibilities:

• Played a key role in the data cleaning process for a large-scale financial data scraping project.

• Utilized advanced SQL and NLP techniques, including lemmatization, to preprocess and refine data collected from 600 websites encompassing 5000 financial funds.

• Ensured accurate and clean data storage, facilitating efficient data retrieval and usage for client-specific applications.

Sign to Speech

Sign to Speech is a project aimed at improving communications with mute and deaf people. Converts sign language to sign and then to speech by text to speech (TTS) technology. 

Responsibilities:

• Designed OpenCV based system to capture signs, Train the model to achieve accuracy.

• Finetuned NLP with RAG to form accurate sentences from letters obtained through Sign language.

• Used TTS for further converting sentences to speech.

Technologies Utilized: Computer Vision, TTS, Lama 2, RAG, SQL, Python 

Sign to Speech

Sign to Speech is a project aimed at improving communications with mute and deaf people. Converts sign language to sign and then to speech by text to speech (TTS) technology. 

Responsibilities:

• Designed OpenCV based system to capture signs, Train the model to achieve accuracy.

• Finetuned NLP with RAG to form accurate sentences from letters obtained through Sign language.

• Used TTS for further converting sentences to speech.

Technologies Utilized: Computer Vision, TTS, Lama 2, RAG, SQL, Python