Hi, I'm Kevin Shah.

Project Speedline (Call Center Process Automation) Analyzed 6 months of AST data to optimize UK call center operations with AI automation opportunities for efficiency. Conducted CXone data analysis to integrate call recordings & addressing billing inaccuracies thereby reducing costs by 35%.

AIRA 360 Member Care (Chat-bot Development) Optimized AIRA 360 for 2 orgs, enhancing IM team's efficiency through pre-checks of customer data using chat-bot. Prompt Testing & feature implementation using Nodejs in backend and implemented Evaluation Log link using React.

Prof. Shouvick Mondal

Utilized LLMs to generate code from prompts to test their sensitivity & robustness to mutations (Radamsa). Evaluation metric: CodeBERTScore & CrystalBLEU | Tested LLMs: Gemini-pro(001, 1.5), GPT-(3.5 turbo, 4, 4-omni). Poster accepted in A* ACM International Conference on Foundations of Software Engineering (FSE 2024)

Tech Stack : GCloud Console | OpenAI Playground | radamsa | hugging face

Pre-joining Assignment: Before officially joining Capillary Technologies, I was tasked with developing an LLM-based Q&A System to answer questions related to the User Documentation of Capillary Tech. I implemented the bot using Langchain and created a demo with Streamlit.

Tech Stack : Streamlit | LangChain | OpenAI (embeddings) | FAISS

It is an advanced platform for data-related courses. Leveraged Google Palm LLM, Langchain, Streamlit, Hugging Face Instructor Embeddings, and FAISS for language understanding, UI, contextual analysis, and efficient database operations. Using a real CSV file as a knowledge base, the system supports human staff in providing quick responses to student queries, fostering a dynamic learning environment through an interactive Streamlit UI.

Tech Stack : Google Palm LLM | Langchain | Streamlit | Hugging Face Instructor Embeddings | FAISS

It is a web application designed for extracting insights from financial news articles. It features URL input for loading articles, LangChain integration for enhanced data understanding, embedding vector construction using OpenAI's embeddings, and FAISS similarity search for efficient information retrieval. Users can interact with ChatGPT for queries and receive answers based on analyzed news articles.

Tech Stack : Streamlit | LangChain | OpenAI (embeddings) | FAISS

Prof. Nipun Batra

Applied BSTS on generative & weather dataset and compared it with benchmark time series models -ARIMA & SARIMA

Tech Stack : ARIMA | SARIMA | statsmodels

Utilized advanced techniques such as Yolov7 for object detection and EasyOCR for text recognition to create Smart ICU Solutions in the healthcare industry. The pipeline's features included k-means clustering and DBSCAN clustering to improve graph digitization.

Tech Stack: Yolov7 | EasyOCR | k-means clustering | DBSCAN clustering

Achieved 1st Place in ML Hackrush-23 challenge. Developed a stock price forecasting project for three subsidiaries of TGD - TGD Consultancy, TGD Automobiles, and TGD Power. Utilized AI and advanced algorithms, employing techniques such as hyperparameter-tuned linear regression, ensemble learning, PCA for dimensionality reduction, and LSTM stacking to optimize profit maximization in the stock market.

Tech Stack: Hyperparameter-tuned Linear Regression | ensemble learning | PCA | LSTM stacking

Designed and implemented a Placement Management System for IIT Gandhinagar, utilizing Flask, HTML, and CSS. Employed MySQL for efficient backend database management, optimizing the placement process.

Tech Stack: Flask | HTML | CSS | MySQL

This project delves into a comprehensive case study on Spotify. Employing SWOT analysis and user-centric frameworks, the initiative focuses on refining market positioning and enhancing social features.
The strategic use of RICE models ensures precise prioritization of innovative requirements over a 3-month period, resulting in a compelling solution roadmap with UI/UX concepts. The project concludes with the establishment of measurable success criteria to gauge the effectiveness of proposed changes.

Build a Sales optimization platform for perishable commodities in Indian food and bakery segment Leveraged advanced analytics for trend analysis, supply-demand prediction, and sales forecasting Benefitted 2 bakeries by improving their sales prediction from 40% to 15% deviation from actuals using Transformer-based architecture for time series forecasting.

Tech Stack: Custom Transformer based architecture | PowerBI for Data Visualization