Highlights:

Skills:Deep Learning, Machine Learning, Large Language Models, LLMs, Computer Vision, Python

Deep Learning Engineer 03/2024 - current
Roll.ai, San Francisco (Remote from Pakistan), United States
Deep Learning Engineer at Roll.ai working AI Editing for Videos
• Architected a multimodal LLM Agent as a centralized AI video editing system, automating social media clip generation, multi-take video orientation/cropping, transcript correction and enhancment, and dynamic VFX/motion graphics creation to streamline post-production workflow. • Optimized real-time video bokeh effects by implementing Monocular Depth Estimation and Human Matting algorithms, achieving 2x performance improvement through parallel processing and batch optimization for live streaming applications • Deployed Active Speaker Detection (LightASD) system for multi-camera live events, automating speaker-camera associations and reducing event video editing time from 48 hours to 1 hour (98% improvement)--
Research Fellow 09/2024 - current
Fatima Fellowship, San Francisco (Remote from Pakistan), United States
Industry: Non-Profit
Research Fellow at Fatima Fellowship
• Working on multi-modal RAG with test time inference for medical domain, advised by Dr. Wei Peng from Stanford University--
AI Fellow 12/2023 - 02/2024
PI School of AI, Rome, Italy
Industry: Education
AI Fellow, selected one of the 10 out of 300 applications
• I was was selected (10 in 300 applicants) from all over the world to turn with fellowship worth 12,500€ where I worked on extracting key information from long context interview transcripts using LLMs (MistralLite, Mixtral8x7B). Worked on RAG using DSPy.--
Machine Learning Engineer 04/2022 - 03/2024
Redbuffer.ai, Islamabad, Pakistan
Industry: Services
Machine Learning Engineer at Redbuffer.ai, worked with several different clients on creating enterprise solutions with Machine Learning.
• Created a fact verification and identification pipeline, integrated multi-threading for a 200% speed increase, fine-tuned LLAMAv2, Dolly-3B, GPT-Neo, and BART for negation generation to search for false news. • Improved OCR latency from 0.6 FPS to 16 FPS on the CPU for streaming devices to blur sensitive information. • Boosted Accuracy to 85% from 30% for logo recognition during cricket matches using CLIP by OpenAI. • Developed a drag and drop RAG system which can be used with any LLM, vector DB, and multi-hop search.--
Software Engineer (Deep Learning and Computer Vision) 12/2021 - 03/2022
Wortel.ai, Islamabad, Pakistan
Industry: Services
SE at Wortel.ai focused on Deep Learning and Computer Vision
• Created a real-time search engine for videos using CLIP by OpenAI, Qdrant for Vector DB, and ElasticSearch. • Performed Multi-label Text Classification using BERT and PyTorch for 1500+ labels.--
Deep Learning and Computer Vision Intern 07/2021 - 12/2021
Wortel.ai, Islamabad, Pakistan
Industry: Services
Deep Learning Intern
• Developed Safety Clothes Detection for Factory workers using Yolov5. • Trained a Celebrity Detection and Classification system using 3DDFAv2, ResNet50, and a 172GB Dataset.--

International Islamic University, Islamabad 08/2018 - 07/2022
Islamabad, Islamabad, Pakistan
Degree: Bachelor's Degree
Major:Computer Science
CGPA:3.43/4.0 Key Courses: Advanced Computer Programming, Object Oriented Analysis and Design, Data Structures and Algorithms, Operating Systems, Database Technologies, Artificial Intelligence, Parallel & Distributed Computing

Login to view resume: Ahmad Resume - Machine Learning, Deep Learning, Computer Vision, Software Engineer, LLMs, NLP