End-to-end AI,
from problem to production.
I'm Harry Hu — an AI/ML Engineer and Data Scientist focused on end-to-end delivery. B.Sc. in Data Science & Mathematics from NYU, currently completing an M.Sc. in Data Analytics & Statistics at Washington University in St. Louis.
I translate ambiguous business goals into measurable systems by starting with the decision, the unit of prediction, the horizon, and the operational constraints. From there, I set a primary metric with guardrails and iterate from interpretable baselines to stronger models.
Problem framing → KPI design → data ingestion → modeling → time-safe evaluation → deployment/monitoring → stakeholder adoption
I own the full loop. I don't hand off a notebook — I ship systems with monitoring, iteration, and stakeholder buy-in baked in. The hardest part is rarely the model; it's clarifying what's being decided, what the right metric is, and what failure looks like in production.
Computer vision is where I go deepest. I've worked across the full CV stack: CNN-based detection (YOLO), multi-object tracking (DeepSORT + Kalman filtering), and transformer-based segmentation (SegFormer, Mask2Former, SAM).
I don't just use these models — I understand when they break. I've diagnosed why Kalman-filtered tracking fails under moving-camera conditions and independently drove the pivot from bounding-box tracking to pixel-level segmentation. I've adapted foundation vision models (MedSAM2) for medical imaging under limited labeled data, and built industrial detection + OCR systems that cut manual workflows by 99%.
I'm a strong believer in vibe coding — using AI agents as force multipliers for engineering velocity. I build with Claude Code, OpenClaw, and AI-assisted workflows daily, treating LLM agents not as autocomplete but as collaborative engineering partners for architecture, implementation, and iteration.
This site itself was built entirely through AI-agent-augmented development — from system design to the interactive canvas background to the glassmorphism design system. I see AI-native development as a core professional skill, not a shortcut.
- GIS + Zillow API valuation: surfaced ~$3M untapped property value opportunity across 220 properties
- Space utilization ML: identified 14% underutilized areas; strategies projected to lift revenue by ~9%
- CV automation: blueprint takeoff 60 min → 30 sec, ~97% accuracy, with imbalance handling + bias monitoring
- Experimentation: A/B tests reduced input time by 25% and cut errors by ~50%
Languages & Frameworks
CV & Deep Learning
ML Techniques
LLM & AI Engineering
MLOps & Engineering
Infra & Tools
AI-Augmented Dev
Open to AI/ML Engineering and Data Science roles in NYC or remote — especially teams where I can own the full loop from problem framing through production deployment, and where AI-augmented development is embraced rather than debated.
Want to see what I've built?
View projectsI'm always interested in thoughtful conversations about machine learning, applied AI, and building useful systems.