Jay Shah

• A chatbot which helps users buy/sell and ask about mutual funds.

• Assisted in Design of Core Engine Architecture, the main service that integrates WhatsApp with various modules for the chatbot.

• Created a reporting module that sends requested reports to investors via WhatsApp. Implemented two cron jobs for automated monthly portfolio updates and capital gains reports 15 days before each quarter. All the reports are stored in AWS S3.

• Developed an AI agent using OpenAI within the reporting module that classifies user queries into the desired report type and extracts the relevant date range to deliver the appropriate time-specific report.

• Fine-tuned a DeBERTa model for multi-label text classification within the reporting module, enabling accurate classification of user queries into report types and date ranges. This optimization significantly reduced costs by Rs 10 per 100 user queries. The model and its dataset are hosted privately on Hugging Face, allowing seamless integration a

• Created 5+ FastAPI’s for LLM interaction. Utilized Google’s Vertex AI to cre‑ate embeddings, semantic matching on queries, and generating responsesfor creating Customer support bot for a client.• ImplementedNeMo guardrails to validate bot responses and execute actionslike sending emails,retrieving ticket details etc.• Built a Retrieval Augmented Generation (RAG) system for a client fromscratch, covering data collection, indexing, retrieval embedding, query pro‑cessing, user input handling, and semantic matching for optimal perfor‑mance.• Created a Document classifier app for a client, enabling classification of up‑loaded PDFs into different entities like Type of Document, Name of Target En‑tity, Date of Document, and Name of Source Entity.