Sameer Ali

I've built production-ready AI systems spanning document intelligence, multimodal video understanding, and LLM-powered applications.

For DocVLM, I fine-tuned Qwen2-VL-2B with QLoRA and 4-bit NF4 quantization to extract structured JSON from invoices, receipts, and complex forms. I delivered the complete pipeline from data preparation and schema validation through inference serving, while reducing memory use by 60%.

I also designed an async Python video AI pipeline combining faster-whisper transcription, OpenCV keyframe extraction, and spaCy semantic embeddings to generate topic-clustered chapter markers. At DevelopersHub, I built NexusStore, an AI-powered e-commerce platform using Groq Cloud LLMs and FastAPI, including an offline semantic fallback for uninterrupted AI features.

I'm focused on bridging AI research and production through Python, FastAPI, TensorFlow, PyTorch, Docker, and dependable model-serving workflows. I'm open to remote AI/ML engineering roles worldwide, especially around LLM deployment, MLOps, multimodal AI, and open-source tooling.