1) Spearheading the development of intelligent LLM-powered agents for SMS marketing campaigns, integrating Twilio for communication, Celery for background task processing, Redis for real-time data management, and MySQL for efficient data storage. Leveraging Python, FastAPI, Node.js, and Llama Index to build scalable and high-performance systems.2) Enhanced user interaction by implementing features like automated calendar invites through Zapier, and leveraging Llama Index to intelligently post-process conversations between the user and AI assistant for improved context and relevance.3) Optimized the backend architecture to efficiently handle high-volume messaging, successfully tested to support 3,000-5,000 SMS per day and process over 100 messages per minute, ensuring scalability and reliability under peak loads.

• Designed and implemented backend architecture using WebSockets, NodeJS, Redis, and ExpressJS to manage real-time audio streams from Twilio, facilitating conversations with a language model (GPT-4) with latency under a second.• Collaborated in designing MySQL databases and cache system to support the application, ensuringefficient handling of 1000 inbound and outbound calls daily.• Utilized Docker to manage multiple NodeJS, FastAPI, MySQL, Redis, RabbitMQ microservices forstreamlining the deployment to stage and production servers.• Engineered a robust queuing system utilizing Celery, RabbitMQ, and Redis to oversee outbound call campaigns for 1000 contacts. This system managed Twilio’s rate limit, retried calls in case of failure due to the rate limit or unanswered calls, processed large documents, and created indexes in the Pinecone vector database for Retrieval Augmented Generation (RAG). It tackled scalability issues, maintained smooth API performance across multiple servers, and resolved time zone conflicts for meeting scheduling with Cal.com.• Developed background tasks for CRM report generation (e.g., Salesforce) by processing audio transcriptions from user-bot conversations, enhancing CRM data utility.

Demonstrated leadership by outlining the roadmap for developing new features, conducting code reviews, offering guidance to junior engineers, and actively contributing to the product.Prompt engineered the GPT-3.5-Turbo LLM to generate summaries of text data, enhancing the creation of short videos. Led a team in the creation of features such as "Video Smart Cropping" a state-of-the-art video cropping solution leveraging object detection and object localization for enhanced saliency. Achieved a significant 30% reduction in processing time and seamlessly implemented the solution on AWS Lambda using containers. Proficiently utilized Python, PyTorch, Fast API, Mediapipe, AWS EC2, and AWS Lambda throughout the role.Implemented an AI camera filter called “Face transformation”. In the second phase of this project, I implemented face similarity capability using “Face Net” embeddings.

Spearheaded the development of an AI camera filter named "Motion Tracking". This feature enables users to create videos where the camera frame dynamically follows their facial movements. This feature was built on Android, it made use of Camera2, Mediapipe, Exoplayer, Android Mediacodec, OpenGL, Kotlin, Kotlin Co-routines and Android custom views. Commended by the co-founder for outmatching his expectation on quality of the feature.

o Implemented and maintained the supporting Android infrastructure for a product called Fovea Stream which runs super resolution algorithms on Digital signal processors (DSP) of Qualcomm mobile processors on videos in real time. Trained and fine-tuned a video super resolution model called ESRGAN to run in real time on mobile devices.o Reduced the AWS costs by 15% by improving the cloud infrastructure. Improved the battery performance of the app by 10% by optimizing the OpenGL rendering.