Pranav Govindaswamy

Developed the Adtua platform using Java-based microservices to connect advertisers with billboard broadcasters, exposing RESTful APIs and role-based access control, and supporting 1,000+ registered users. Integrated a natural-language driven search capability using Python services with Vertex AI for intent extraction, converting free-text advertiser input into structured query parameters and reducing average inventory selection time by 40%. Designed a scalable inventory service backed by indexed database queries, displaying 500+ advertising spaces and supporting filters for location and display type under 5 seconds. Streamlined backend services for campaign creation, messaging, and listing verification using asynchronous workflows and validation layers, supporting 300+ advertiser–broadcaster conversations.

Deployed LLM-based APIs using Amazon Bedrock (Claude, Titan, Cohere), supporting 3,000+ monthly AI requests across production applications for summarization, question answering, and decision-support workflows. Implemented retrieval-augmented generation (RAG) pipelines using Amazon OpenSearch and S3, indexing approximately 8,000 documents and improving response relevance and factual grounding by 30% compared to prompt-only approaches. Executed AI data preparation and training workflows using Vertex AI, processing 6,000+ records per training cycle for feature preprocessing, and evaluation prior to downstream deployment in production inference systems. Developed serverless GenAI workflows using AWS Lambda and DynamoDB to manage prompt metadata, versioning, and fallback routing, supporting 10+ active prompt versions while reducing inference failures by 50%. Built AI usage and cost dashboards using Amazon QuickSight, tracking weekly model usage, error rates, and cost trends and enabling a 20% reduction in GenAI operational costs through usage optimization.