At Dresma AI, I architected a GPU-accelerated Background Removal API using TensorRT-optimised PyTorch, reducing p95 inference latency from 850 ms to 510 ms. I also designed Flask and FastAPI microservices on Google Cloud Run.
I designed the Brand Intelligence System, an event-driven pipeline that syncs Amazon A+ content metadata from MongoDB to a Neo4j knowledge graph. Its two-phase publish protocol eliminated race conditions without polling or scheduler downtime.
At Deepmindz Innovations, I led GenAI initiatives for face swapping, lip-sync, and voice cloning, and fine-tuned multilingual LLM assistants using RLHF and prompt engineering. I also built skin analysis models and a sentiment classification and review summarisation system that reduced manual review time by 60%.
My projects include a RAG-powered enterprise knowledge bot using LangChain, Pinecone, and GPT-4, and a deepfake detection pipeline that achieved 94% precision on benchmark datasets. I work across computer vision, generative AI, and cloud-based ML systems.
