Single handedly developed entire backend and infrastructure of a solution consutlant application using advanced RAG

Developed an on premise RAG application end to end using AWS postgres, OpenAI , FastAPI and OAuth2

• Led the development of time series predictive maintenance algorithms for cloud infrastructure efficiency

• -Engineered advanced predictive solutions with 0.85 f1 score utilizing state space models and Data-as-a-Model algorithms, resulting in an annual cost reduction of approximately $6,000 on AWS servers

• Developed FCW algorithms with an approval rating of 85 percent using MobileNet and sohisticated postprocessing

• Fully Resposible for development of Mindcase's legal copilot system, now in pilot across Gujarat and Delhi law firms.

• Managed backend pipeline for Mindcase's legal copilot, enhancing functionality and reliability.

• Led OpenAI LLM's model steering using multihop adaptive Retrieval Augmented Generation, boosting performance.

• Optimized Pinecone and Supabase data retrieval, significantly enhancing user experience.

• Developed robust backend APIs with FastAPI, increasing system interoperability and efficiency.

• Directed large-scale legal web scraping with Selenium, gathering critical data for RAG.

• Contributed to Mindcase projects using locally deployed LLMs using Ollama, Streamlit/Chainlit on GCP and Azure.

• Custom Web Chatbot Development and Dynamic Content Recommendation System:

• -Designed and implemented a custom chatbot solution from scratch, leveraging Crawlee for web crawling, Selenium for dynamic content extraction, and Beautiful Soup (bs4) for parsing HTML.

• Developed an user interface using Streamlit and robust backend APIs with FastAPI,

• Utilized Chomadb to create a recommendation system based on crawled content from any given website

• Seamlessly integrated the api, providing personalized recommendations for products, services, or information

• Leveraged computer vision to build a multimodal drone detection MVP and deployed Flask app on Azure Server

• Fine-tuned YOLOv8 models, achieving an impressive mAP of 0.68 on challenging drone datasets.

• Curated and preprocessed diverse dataset from Roboflow, ensuring model adaptability to real-world scenarios.

• Applied FFT based signal processing on RF signals, attaining a high F1 score of 0.95 for accurate detection

• Conducted extensive research in Drone Forensics, delivering a successful MVP for a sota multimodal drone detection