Syed Anwar

Collected, annotated, and cleaned vehicle imagery to create a training dataset for detection. Optimized a YOLOv8 model to improve object detection accuracy for counting use cases. Applied model training and tuning steps with real-time deployment constraints in mind using edge hardware.

Built a facial-recognition attendance system using classical ML classifiers for individual recognition. Designed the system for secure deployment within a department environment to improve both efficiency and security. Used an end-to-end workflow that maps captured faces to stored identities and records attendance.

Fine-tuned YOLOv8 and sequence models to detect sign language for a government initiative. Integrated the detection pipeline with ChatGPT-style language processing to improve downstream understanding. Combined computer vision and language components to form an end-to-end sign language solution.

Created an AI system that detects contextual and translation-related errors by comparing source and target language documents using GPT-based prompting. Added tooling for text formatting and adaptation, including tone changes and rephrasing. Collaborated with backend development to integrate CRUD functionality via REST APIs.

Built a transformer-based text generation model (about 5M parameters) using GPT-4o and PyTorch, then evaluated it on task-specific datasets. Developed a transformer-based code completion model for Python and Java (about 19M parameters) and compared results across runs. Focused on training and evaluation workflows for compact, task-aligned architectures.

Built an AI-powered medical coding platform that extracts clinical data from PDF and image documents and generates structured diagnosis and procedure codes. Implemented OCR-based workflows, ICD hierarchy lookup, validation across multiple documents, and bulk CSV/XLSX code generation. Developed secure admin APIs, approval flows, and usage-based billing logic, integrating Stripe for real-time payments.

Created and fine-tuned a persona-driven crypto Twitter agent by scraping, cleaning, and adapting a Llama model to match an existing agent’s style. Shipped a gated terminal combining AI signal generation, crypto sentiment, and technical analysis using live market data. Built RAG-based context enrichment and an AI receptionist workflow with n8n, ElevenLabs, and speech transcription, optimizing response latency to about 850ms.

Delivered computer vision, deep learning, and NLP systems, including training Transformer-based models and deploying AI agents. Trained transformer text generation and code completion models from scratch using PyTorch and GPT-4o as part of evaluation-driven iterations. Built a real-time pose estimation system and a facial-recognition attendance flow with PPE compliance alerts. Developed an AI job-replacement risk and skills gap agent using LangGraph and supported business decisions with ML-based data analysis.

Developed computer vision, deep learning, and LLM-based chatbot and automation systems for business use cases. Built an AI web app to detect translation errors by comparing multilingual documents against English source text using prompt engineering and token-cost control. Implemented an LLM knowledge assistant with RAG over scraped content, plus retrieval improvements using query expansion and re-ranking. Trained and deployed YOLO-based models for sign language detection, created time-series forecasting pipelines with Prophet, deployed ML pipelines on AWS services, and produced dashboards in Power BI.

Led a small team to build a facial-recognition smart attendance system that replaced a manual fingerprint-based process and reached 94% accuracy. Contributed to a smart vehicle counting system using machine learning for automated detection. Supported government infrastructure integration efforts as part of an ICT Developer fellow role.

Analyzed booking records to identify patterns behind a 27% ticket cancellation rate and support targeted mitigation strategies. Built predictive models to forecast operational trends and improve performance. Communicated findings into actionable recommendations that informed business decisions and operational planning.

Trained and evaluated a gender prediction model under PhD supervision, achieving high accuracy. Implemented and compared multiple detection architectures (including YOLOv8) for license plate recognition and supported an academic publication effort. Focused on model evaluation and experimentation for practical computer vision outcomes.