CODE:751 Product 3 cases Nominal
Product

Product

Three systems I owned end to end — content-safety auto-labeling at Pyler, defect synthesis and steel-plate OCR at Aiv. Each reads the same way: the problem, what was mine, and what moved. One of them started as research and I carried it onto the line myself rather than handing it over.

Case 01/03
Pyler · Product · Content SafetySystem I own

Mining Decision Trees from VLM Reasoning

Content-safety scoring — 5 categories × severity — drifted between labelers. The written guidelines were flat ladder tables, but the decision logic that actually produced a score lived unwritten in each labeler's head, so the same clip scored differently depending on who saw it.

category signal? exemption gate? severe? meaningful? weak? no signal exempt severe meaningful weak else 0 0 5 3 1 0

Signal check → exemption gate → severity ladder. Every score traces to a path.

End-to-end ownership: problem framing, the mining pipeline, the tree design, and the prompts now in production.

Macro F1 0.777 → 0.857 (+0.080), and over 4× the labeling throughput of the process it replaced. Every score is auditable now: you can point at the node that fired.

My contribution
  • The insight. A VLM that explains its score leaks the decision path — mine enough traces and the latent tree is recoverable.
  • A pipeline over 299K traces. Group into 30 buckets (5 categories × 6 scores), a small LLM extracts the decisive signals, another synthesizes them into a yes/no tree (YAML) injected back into the prompt.
  • Engineered the trees. Merged equivalent signals into single axes, pulled exceptions to the front as early gates, and collapsed the fuzzy 2-vs-3 / 4-vs-5 bands into a stable 0 / 1 / 3 / 5 scale.
Macro F1 +0.080Throughput 4×+Precision +0.103Recall +0.050299K tracesYAML trees
Case 02/03
Aiv · Product · Industrial ADProduct I own

Background-Aware Defect Synthesis

Real defects are scarce on a manufacturing line — too few, and too narrow in type, to train a detector that generalizes. So I generate them.

Mine end to end: first author on the method, then the pipeline and rollout that put it on the line — research carried into production, not handed off.

Best generation quality (FID / LPIPS) over DFMGAN & AnomalyDiffusion on MVTec-AD & LOCO; raised detector precision & recall on the real production line.

My contribution
  • Disentangled defect from background. Diffusion with masked cross-attention and separate defect/background context vectors, trained with a disentanglement loss so a synthetic defect respects its surroundings.
  • Controlled placement at inference. DDIM inversion on a normal image, then sample the defect onto a refined mask region.
  • Made it affordable on one GPU. Flash-Attention + DeepSpeed + FP16, wrapped in an MLOps pipeline.
PyTorchDiffusersFlash-AttentionDeepSpeedFP16
Case 03/03
Aiv · Product · OCRProduct I own

Real-Time OCR Pipeline for Steel-Plate IDs

Read identifiers stamped on steel plates outdoors — from CCTV and mobile — fast and reliably, under any weather.

Mine end to end: pipeline design, the models, profiling and optimization, and the rollout. A product I owned, not a component I contributed to.

Client confirmed throughput and recognition accuracy both rose sharply after rollout.

My contribution
  • Two-stage pipeline on NVIDIA Triton. Detection crops the region of interest; a recognition model reads it and matches the DB.
  • Profiled, then cut runtime. Triton logs showed model runtime was the bottleneck, not pre/post — so ONNX → TensorRT, warmup, and CUDA graphs.
  • Made it weather-proof. Test-time augmentation (rotation, padding) baked into the pipeline.
TritonTensorRTONNXCUDA GraphsDetection + Recognition