
Your backend is the bottleneck. I fix that.
40min → 1min import · 108× faster response times · APIs that scale
Deep dives into specific technical problems — what broke, how I diagnosed it, and what changed.
Started as a computer vision experiment, ended as a production API with sub-2s inference across 20K+ facial records. The technical decisions were sound. The architecture wasn't.
20K+
Facial records
737 ms
Fastest endpoint
FaceNet512
Model
A ~100MB automotive inventory import was silently hanging for 40 minutes. I diagnosed four root causes and restructured the pipeline — cutting it to under 1 minute.
40 min
Before
~1 min
After
97% faster
Improvement
Two identical APIs, same 8.8M row dataset, different query plans. The result: 666× faster on single-record lookups, 108× on list queries — without touching the server.
8.8M rows
Dataset
666×
Best speedup
99.85%
Latency reduction
A benchmark across 4 approaches — row-by-row, batch 1K, batch 10K, and Go goroutines — on a 1GB VM. The difference between the worst and best: DNF vs 20 seconds.
1M rows
Dataset
20s
Best time
3m 19s
Worst (completed)
Production systems demonstrating performance optimization and scalability.
Live
CompletedA collection of personal experiments where I explore ideas, prototype concepts, and test new technologies, including AI, computer vision, and real-time systems.
Complete
Live
Live
Completed
Completed
Completed
Completed
Completed
Completed