PAPER — 01
Optimizing EVEv2 Connector with CUDA Stream Concurrency in VILA-U
A. Mateen, M. N. Ali
2026 · UNPUBLISHED
I build production AI systems — RAG assistants, vision-language models, CUDA kernels. BS Artificial Intelligence @ FAST NUCES, shipping real software from Islamabad to the world.
Production-minded AI systems — each one deployed, measured, and documented. Metrics are from real runs, not aspirations.
RAG-powered hotel front-desk assistant with real-time voice interaction. Retrieval pipeline over property knowledge, FastAPI backend, Qwen LLM with streaming responses.
Rewrote VILA-U's EVEv2 connector with CUDA stream concurrency — overlapping compute and transfer to kill pipeline bubbles in vision-language inference.
Multi-scale region–phrase alignment framework for remote-sensing VLMs. Grounded language to satellite imagery across scales for retrieval and captioning.
Fake-news detection with hybrid Bi-LSTM + handcrafted linguistic features over 72K articles. Clean data pipeline, reproducible training, honest evaluation.
Browser extension for real-time video subtitle translation. Faster-Whisper streaming transcription piped through WebSockets with sub-second latency.
Mobile accessibility app: point the camera at any text, hear it read aloud. On-device OCR plus natural TTS, built for low-vision users.
Internships, teaching, and research — each one left something running in production or in print.
Built backend systems for production AI workloads — including metering and billing infrastructure for an AI capstone. Learned how AI products actually make money.
Applied vision-language AI to cultural heritage preservation — grounding modern VLMs in the messy reality of historical artifacts and sites.
Guided Operating Systems labs for undergrads — debugging, evaluation, and explaining concurrency until it clicked. Teaching is the fastest way to learn.
Deep learning, computer vision, NLP, systems. The degree taught theory; the projects taught everything else.
Three manuscripts in 2026 — systems optimization, vision-language alignment, and applied NLP. Unpublished, under review.
The technologies I reach for when something needs to work in production, not just in a notebook.
I'm looking for applied AI engineering roles — teams shipping LLM products, RAG systems, or anything where models meet real users. My inbox is open.