<Cyrus Mante>

build and ship generative AI proof-of-concepts that solve real business problems — covering agentic workflows, prompt engineering, and cloud deployment within a cross-functional CoE.

write clean, tested Python, conduct code reviews, and integrate no-code tools — keeping delivery fast without cutting corners on reliability or security.

built internal LLM-powered web apps that let developers and managers interact with AI tools without needing to write a single prompt themselves.

proposed, documented, and deployed AI tools in collaboration with onshore teams — from idea to production with actual adoption.

found my own clients and shipped computer vision projects for small businesses in logistics, security, education, and finance — owning each one end-to-end, from figuring out what they actually needed to running it in production.

built a face recognition attendance system on OpenCV and facial embeddings — with liveness detection so a photo couldn't clock you in — backed by SQLite and a Streamlit dashboard for the people who had to read the reports.

put license plate recognition on the edge for automated parking billing: YOLO for detection, quantized with TensorRT for ~2x faster inference on an ESP32, EasyOCR for reading the plate, and fees computed server-side.

digitized receipts for an accounting firm — OCR text cleaned up by Gemini, wrapped in Flutter so it ran on both platforms — and built a webcam-based OMR grader for a school that segments bubble grids and pulls the answer key from a template.

built a receipt scanner app in Python + React Native, revamped a car plate detection model improving efficiency by ~40%, and taught an AI workshop as the first AI intern at Cryptex.

also did early ML work at Dashlabs and AIA — cleaning healthcare data, building a heart disease prediction model from ECG data, and developing a next-best-offer model that improved conversions by ~20%.