HELLO I'M
HAMZA AZIZ
AI PRODUCT ENGINEER
SUMMARY
AI Product Engineer with 3 years of experience in Computer Vision, Conversational/Generative AI, full-stack LLM applications, and applied MLOps. Built 4 production-grade facial recognition systems across education and enterprise sectors, reducing manual effort by 90% with >99.99% accuracy via real-time pipelines and automated Milvus DB updates. Currently shipping a real-time, multi-tenant DriveThru voice ordering agent on FastAPI + Socket.IO with a LangGraph (ReAct) agent over Google Gemini, alongside earlier Rasa-based voice-to-SQL assistants. Comfortable shipping LLM applications end-to-end — FastAPI + Next.js services with retrieval-augmented, citation-grounded narrators over vLLM / OpenAI-compatible endpoints. Proficient in GPU inference (Triton, TensorRT, DeepStream) and CI/CD deployments.
SKILLS
- Python
- Tensorflow
- PyTorch
- TensorRT
- Milvus
- PostgreSQL
- MongoDB
- Redis
- Neo4j
- MinIO / S3
- LangChain
- LangGraph
- Agentic AI (ReAct)
- Function Calling / Tool Use
- Rasa
- Gemini
- vLLM
- RAG
- NLP
- LLM
- Prompt Engineering
- FastAPI
- Socket.IO
- Pydantic
- Next.js
- TypeScript
- Docker
- Kafka
EDUCATION
M.Tech (Executive) in Artificial Intelligence and Machine Learning
Bachelors of Technology in Computer Science and Engineering
Senior School Certificate Examination
Secondary School Examination
WORK EXPERIENCE
AI Product Engineer
Shipping production features for a real-time, multi-tenant DriveThru voice ordering agent. Contributed to the modernization from a Rasa-based NLU platform to an agentic LangGraph + Google Gemini stack — working on entity extraction, conversation flow, and tool-use orchestration. Helped lift live client success metrics from ~25–30% to 60%+ across 4 POS integrations.
Assistant Engineer - AI
Contributed to Conversational AI project for College SIM Portal; integrated Rasa, Gemini, and API for real-time data retrieval, boosting user satisfaction by 25%. Developed a Multi-Person Facial Recognition System, saving 100+ admin hours monthly
AI Intern
Streamlined development of Facial Recognition System; reduced data errors by 50%. Contributed to training sessions on Computer Vision projects, assisting students in leveraging YOLO for enhanced object detection.
PROJECTS
DriveThru Voice Ordering Modernization
Real-time, multi-tenant DriveThru voice ordering agent on FastAPI + Socket.IO with a LangGraph (ReAct) agent over Google Gemini 2.5-flash (with 2.0-flash fallback), Redis-backed session state, and MongoDB-backed order-event audit trail. Hardened entity extraction against ASR disfluencies, added combo/multi-size guard logic, and integrated 4 POS providers (QUBeyond, Clover, Speedline, Deliverect) with Jinja2-templated per-restaurant prompts.
HalalLens — Halal Investment Research Copilot
Full-stack Shariah-compliance research platform for Indian equities — FastAPI (33 endpoints), Next.js 14 + TypeScript, PostgreSQL, Docker, GitHub Actions CI. Built an AAOIFI-style ratio engine across 5 compliance screens and a dual-mode “Explain Why” narrator (rule-based + vLLM / OpenAI-compatible LLM, citation-grounded). In-process workers crawl BSE & NSE announcements, extract financials from PDF/XBRL, and refresh compliance verdicts on a schedule.
ViewG6 Voice Assistant (Conversational AI for College ERP)
Developed a Conversational AI system for Global InfoVentures Pvt. Ltd., enabling voice and text-based queries to retrieve real-time data from the SIM Portal. The project was integrated with the G5 Portal to enhance user interaction across platforms.
CCTV-Based Face Attendance System
Launched on NVIDIA DGX A100, orchestrating 20+ live CCTV streams with TensorRT-optimized SCRFD (scrfd_10g_gnkps) and w600k_r50 models. Pioneered dual dataset staging (mobile & CCTV) over Milvus Vector DB to mitigate domain shift, achieving >99.99% face recognition accuracy across varied environments. Architected a CI/CD-style data pipeline that extracts faces from elevated-angle surveillance cameras, filters best-quality faces per person, and auto-updates the vector database — improving recognition rates by 25% in practical deployment.
PTZ-Classroom Attendance System
Automated classroom attendance by deploying rotating PTZ cameras integrated with the face recognition pipeline — eliminating manual roll calls and reducing proxy attendance by 25%.
Edge-Based Office Attendance System
Commissioned a Jetson Orin Nano for real-time on-device inference using DeepStream SDK, a Raspberry Pi AI camera, and proximity sensors. Eliminated manual attendance logging and replaced legacy biometric systems with 100% contactless AI-driven tracking, enhancing hygiene and operational efficiency in office setups.