• Developing a real-time telephony system with personal AI using FastAPI, integrating NVIDIA Riva for speech-to-text and text-to-speech conversion.

Led the quantization of LLaMA3 and Personal Language Conversational AI Models using Qualcomm AI Stack and AIMET, preserving model accuracy for On-Device AI inference.Created an integration C++ API for the loading and execution of quantized LLM model binaries, for inference on Windows and Android edge devices.Integrated with FastAPI by creating a Python wrapper, enabling seamless deployment of the LLM as an application on the Windows platform.

Data Science/Machine Learning Platform (DSMLP)• Update MTL-Attendance for student attendance tracking at UCSD, utilizing Python, AWS Lambda, S3, and Terraform.• Redesigned the Cluster and Pods Status dashboard using React.js, enhancing real-time monitoring of GPU and CPU resources, resulting in a 25% reduction in troubleshooting time.• Developed Airflow DAGs to fetch enrollment data, and manage user drops and adds, enhancing class roster management for over 150 courses.• Enhanced Canvas middleware integration with course management systems by implementing CI/CD regression testing pipelines, resulting in 3% faster enrollment processing and improved deployment stability.• Worked on spinning up Jupyter Notebook environment with GPU support on AWS EKS containerizing JupyterHub instance using Docker and Kubernetes, enabling the on-demand creation of Jupyter instances.

• Released an LLM-based Retrieval-Augmented Generation (RAG) medical chatbot, leveraging Langchain and LlamaIndex, enabling faster and more efficient access to data and information from medical publications for researchers.• Fine-tuned the Llama3-8B model using LoRA resulting in a 15% increase in response accuracy improving the chatbot’s ability to provide precise medical information.

• Collaborated with the E-commerce service team to develop backend microservices using Python, Django, and Flask, directly contributing to an annual revenue increase of $500k.• Implemented a Customer onboarding microservice that led to a 20% increase in user registrations in Q1 2023. Enhanced user engagement through personalized onboarding experiences.• Developed an inventory management microservice incorporating real-time updates via Kafka, Lambda, and EventBridge, reducing inventory sync lag by 30% and minimizing stock discrepancies.• Created a delivery tracking microservice with Google Maps integration, which decreased delivery time delays by 8%.• Wrote SQL queries in PostgreSQL for supporting business operations including inventory management, and sales analytics• Built a Backend For Frontend (BFF) service using Node.js, which reduced API response times by 20% increasing the scalability of front-end applications.• Deployed LSTM models using PyTorch for future sales forecasting, achieving an 85% accuracy rate to generate actionable insights, enhancing the reliability of sales predictions, and improving decision-making.• Revamped the software development life cycle by implementing Agile methodologies; accelerated product release cycles by 40% and improved team collaboration.Leveraged collaborative filtering and content-based recommendation systems to personalize restaurant and productrecommendations, resulting in a 25% increase in order frequency and customer engagement.• Developed customer churn prediction models using historical transaction data and behavioral patterns, enabling proactiveretention strategies that reduced churn by 20%.• Implemented A/B testing frameworks to evaluate recommendation system performance, resulting in a 15% improvement in model accuracy and business metrics

• Co-authored published two papers on the topics• A Neural Network Model to Predict the Radiation Resistance of Dipole Antenna. (https://ieeexplore.ieee.org/document/9847856)• A Neural Network Model for Effective Dielectric Constant Prediction of a Two-Layered Microstrip TransmissionLine (https://ieeexplore.ieee.org/document/9848322)• Developed Deep Learning solutions for antenna and transmission line design challenges.• Calculated the ideal values for the parameters using Python, TensorFlow, PyTorch, Keras, and TensorFlow.js

Worked under the project, Melioration of IEEE802.15.4 wireless communication standard under avionics department in VSSC