AI undergraduate building real systems — LLM-powered chatbots, RAG pipelines, CV tools, and CUDA inference optimization.
I'm an AI undergrad at FAST NUCES Islamabad — not just reading papers but implementing them, breaking them, and improving them. I care about the full stack: from model architecture down to APIs and real products.
My work spans CUDA-level inference optimization on Vision-Language Models, voice-interactive RAG chatbots, and real-time subtitle translation Chrome extensions. I recently served as a Lab Demonstrator at FAST, debugging and guiding students through AI coursework.
FastAPI-based AI assistant using Qwen LLM with RAG and CRM features. Full real-time voice pipeline with Moonshine ASR + Kokoro TTS. Connected to Calendar, Calculator, and OpenWeather APIs for live hotel operations.
Chrome extension for real-time subtitle generation and multilingual translation of streaming content. Low-latency pipeline using Faster-Whisper, WebSockets, and sliding-window audio segmentation.
Multi-stream parallelism on EVEv2.0's dual-branch FFN. Zero retraining, zero GPU memory overhead.
Extended DGTRS-CLIP with dual-encoder contrastive learning. Benchmarked with Recall@K across I2T and T2I directions.
Mobile app combining OCR text detection and TTS to read signs, menus, and documents aloud in real-world environments.
Open to internships, research collabs, and interesting AI projects. If you're building something worth building — let's talk.