Building AI
that ships to
production
Things I've shipped.
Company Projects
Computer Vision: PPE & Zones
COMPANYEdge CV safety system
Custom YOLO models on OAK edge cameras for PPE and zone detection.
Healthcare Reporting Automation
COMPANYEnterprise reporting automation
A C#/.NET Excel add-in that cuts reporting for 60+ hospitals from hours to minutes.
Personal Projects
Lifestyle Tracker
PERSONALWeight-cut tracker that shows its working
An Android weight-cut app that never states what it cannot measure.
Tulu→Kannada Script Mapper
PERSONALHandwritten OCR capstone
A capstone OCR system mapping handwritten Tulu characters to Kannada.
The stack, end to end.
AI / ML
Custom-trained detection models and CV pipelines, from dataset to deployed inference.
LLM / RAG
Retrieval-grounded assistants: embeddings, vector search, and streaming generation.
Optimization & Serving
Getting models fast enough to run in real time — on edge devices and GPUs alike.
Languages
Production code across backend services, desktop add-ins, and the web.
Backend & Web
Backend APIs and product surfaces — services, authentication, dashboards, and web applications.
Cloud & DevOps
Deploying and operating AI workloads on managed cloud and GPU infrastructure.
Databases
Relational stores, caches, and managed backends behind the products I ship.
How I got here.
- 2026
AI Engineering & Mobile ML
Built Codeward (RAG interview platform) and FolliScan, training custom MobileNet models and deploying TensorFlow Lite for fully on-device inference.
- 2025
Document AI at Scale
Shipped intake→parsing→analytics (FastAPI + Postgres), tuned latency < 100ms.
- 2024
Azure AI Certified
Earned AI-900 & AI-102, Built Excel Add-in for Automating Zero Balance and AR Reports saving $120k.
- 2023
Custom YOLO Pipelines
Trained YOLO models for PPE & gesture detection; deployed inference services.
- 2022
Started AI/ML projects
Built first CV prototypes and automation tools while learning PyTorch/TensorFlow.
What broke, and why.
API Authentication Failure — Root-Cause Debugging
Problem: API calls kept failing even though the token, method and endpoint were correct.What failed: Standard debugging on the code path and headers didn't show anything wrong.Breakthrough: After pairing with a senior, I discovered that the local machine was missing a required certificate — the same calls worked on a VM with the cert installed.Lesson: Debug beyond the obvious code suspects; environment and system dependencies can be the real root cause.
Factory Long-Range Detection on OAK-D
Problem: A YOLO-based people detector struggled to identify workers far from the camera on a factory floor, with accuracy dropping below 40%.What failed: Using generic people-detection datasets didn't generalize to long-range, small-person scenarios on the actual OAK-D footage.Breakthrough: I built a custom dataset from similar factory environments, annotated new frames and retrained the model, pushing far-range detection accuracy above 85%.Lesson: When a suitable dataset doesn't exist, create one — targeted data + custom training often beats generic pretrained models, especially in niche CV use cases.
Certified on Azure AI.
Azure AI Engineer Associate
AI-102Designing and implementing Azure AI solutions across Vision, Language and Search.
Azure AI Fundamentals
AI-900Fundamentals of AI/ML on Azure, responsible AI, and core cognitive services.