Browse categories
Explore
Fiverr Pro
English
$
USD
Hamza Aziz
@hamzaziizzzz
AI Engineer for Voice AI Agents, LLM Apps and Computer Vision
India
English, Hindi
About me
Hi, I'm Hamza, an AI Engineer with 3+ years of full-time experience building voice agents, LLM apps and computer vision systems that run in front of real users.
What I build:
- Voice AI: real-time drive-thru ordering agents (FastAPI, Gemini) with deterministic guardrails
- LLM apps: retrieval over your data, structured outputs, agentic workflows with human approval
- Computer vision: multi-camera face recognition on DGX A100, RTX 4090 and Jetson Orin Nano, 99.99%+ accuracy in controlled UAT
Let's build something that works in production. Reach out!
Skills
Hamza Aziz
Offline •
See my services
AI Websites & Software
I will build a real time face recognition attendance system for cctv cameras
AI Integrations
I will build real time video analytics with yolo object detection and tracking
Portfolio
Work experience
AI Product Engineer
VoicePlug Inc. • Full-time
Aug 2025 - Present • 1 yr 3 mos
Real-time, multi-tenant drive-thru voice ordering agent: a FastAPI and SocketIO service orchestrating Google Gemini on Vertex AI. - Contributed to the rebuild from a Rasa intent-and-forms bot into an LLM-first Gemini agent, replacing NLU training data and per-item forms with prompt- and menu-driven ordering. - Order agent on a cached system prompt; a second Gemini model with structured JSON output turns the agent's spoken confirmations into cart updates. - Deterministic Python guardrails around both models, and an async, Redis-backed per-session cart pipeline. - Built code guards for combos and sizes, first as Rasa training data and entity-extraction fixes, then as guards in the Gemini agent. - Hardened ordering against speech-recognition errors: misheard sizes and quantities, items said after filler turns, filler and apology handling. - Built the staff-handoff behaviour, the pickup fast path, a hold guard for items named mid-question, and Spanish-language handling. - Worked on per-restaurant prompt configuration and POS integrations with product, support and QA.
Assistant Engineer - AI
Global Infoventures Pvt. Lyd. • Full-time
Jul 2023 - Aug 2025 • 2 yrs 1 mo
- Solo developer of a CCTV face recognition attendance system (InsightFace-REST with TensorRT, Milvus, PostgreSQL), delivered as 4 CCTV, PTZ and edge variants across DGX A100, RTX 4090 and Jetson Orin Nano; over 99.99% accuracy against the attendance register in controlled UAT at the main and library gates. - Led G6, a voice assistant for a college ERP across 4 institutes: Rasa routing, Milvus schema retrieval, and Gemini emitting query fields and conditions for application-built ERP requests rather than generating full SQL; near-100% intent classification in manual testing with 50+ voice variations per module. - Mitigated domain shift in facial recognition using a CI/CD-style pipeline combining mobile and CCTV datasets, improving cross-domain recognition by 14%. - Automated nightly face-vector updates from 8 ft-high CCTV, raising real-world accuracy from 85% to 99%. - Integrated PostgreSQL for recognition logs and built an API sync module with the college ERP, enabling real-time attendance updates and ERP-driven alerts at a main gate averaging ~2,000 entries a day.
