Omima Shah

I've built AI/ML and full-stack products including Awaaz-e-Gum, a real-time Pakistani Sign Language platform that converts live gestures into text and synthesized speech.

For Awaaz-e-Gum, I co-developed an RTMPose 133-keypoint pipeline, Mamba2 classifier, FastAPI inference services, and a Next.js interface, reaching approximately 80% mean accuracy across 65 gesture classes and validating the system with 86 unit tests.

I've also built a clinical RAG prototype using LangChain, Pinecone, and MIMIC-IV-Ext data; fine-tuned Qwen2-VL-2B-Instruct with 4-bit QLoRA for document-to-Markdown generation; and implemented vision transformer, masked autoencoder, and diffusion-model projects from scratch in PyTorch.

At EDGE C Pvt. Ltd., I coordinate AI/ML, software, and data-intelligence initiatives across a cross-functional team, translating stakeholder needs into technical scope, data workflows, model/API integrations, testing, and delivery plans.