Jay Shah

Engineered a chatbot facilitating mutual fund transactions and inquiries via WhatsApp using WhatsApp templates and Twilio. Designed the Core Engine Architecture, integrating WhatsApp with various chatbot modules. Developed a reporting module using FastAPI and OpenAI agents for intent classification, automating report delivery via WhatsApp and AWS S3. Implemented 2 cron jobs for sending reports monthly. Created from scratch Purchase Module using FastAPI, integrating it with the Core Engine for seamless mutual fund transactions. Fine-tuned a DistillBERT model for multi-label classification for reporting, achieving 99% accuracy, reducing inference latency by 80% and reducing costs by Rs 10 per 100 queries. Deployed the model on AWS SageMaker Serverless Inference with 0$ cost. Deployed applications on AWS EKS using GitHub Actions and Kubernetes YAML files.

Hired and managed two interns within a small budget, conducting weekly standups, assigning tasks, and ensuring project alignment. Led the MVP development for Qlink, building the chatbot from scratch on WhatsApp using Gupshup and custom flows. Designed and implemented the backend architecture and end-to-end conversation design for the MVP. Built a CI/CD pipeline using GitHub Actions, building Docker images and pushing to Vultr Container Registry and then deploying to Vultr EC2 Instance with a cost of $14/month. Applied Design Patterns and SOLID principles, implementing Singleton Logger, Template Response Manager, and Abstract Patterns.

Developed 5 Customer Support chatbots using RAG systems and Microsoft AutoGen, covering data collection, embedding generation, and similarity search optimization. Built a chatbot for Rackspace using FastAPI for LLM interaction, leveraging Google Vertex AI for embeddings and Nvidia NeMo guardrails for response validation. Created a document classifier application to categorize uploaded PDFs into structured entities. Optimized Casino data using SQL in Google BigQuery for Google Recommendation AI, training recommendation models like “Recommended for you” and “Similar items”. Integrated Google Retail API into their web app. Automated email summaries using Pipedrive API, increasing lead follow-up speed by at least 20% for sales team.

Enhanced refund tracking for Byju’s by implementing refund timeline visibility using ReactJS and Flask APIs. Developed 10+ Python Flask APIs for the Ops Panel, automating manual tasks and reducing workload by 50%. Implemented Slack alerts in cron jobs for real-time issue notifications in daily files sent to banks. Built a Java-based SFTP solution for secure and efficient uploads to the Bank of Baroda server. Addressed production issues and solved them promptly.