Services
AI/ML
We keep this practical: small pilots that prove value, then expand. We focus on edge vision and document/log intelligence — with private/on-prem options.
Edge Vision (On-device)
- Object/zone events and counting.
- Lightweight visual classifiers for on-device use.
- Camera & lighting setup; per-device calibration.
Document & Log Intelligence
- Spec/RTL/log Q&A scoped to your repos.
- Template drafting (test attributes, checklists) with human approval.
- Privacy: on-prem options, redaction, role-based access, audit trails.
MLOps & Delivery
- Clear acceptance metrics (precision/recall, latency, false-alarm rate).
- On-prem/edge packaging; offline-first sync; simple dashboards.
- Handover docs so your staff can run it without us.
How we start
- Send sample photos/logs or a short site video and constraints.
- We propose a 4–6 week pilot with fixed scope and pass/fail criteria.
- If it works, expand by station/line and train your team.
Detailed Capabilities
Model Development & Evaluation
- Data plan: labeling, QA, sampling; safe augmentation strategy.
- Baselines: HF models (YOLOv7/RT-DETR) + classical checks.
- Metrics: precision/recall, F1, latency, false-alarm rates; ablations.
Edge AI Optimization & Accelerator Integration
- Quantization (INT8/FP16), pruning; runtimes: TensorRT, ONNX Runtime, OpenVINO.
- Targets: Jetson, x86 GPU, ARM SBCs; containerized builds.
- Throughput tuning within power/thermal envelopes.
Vision Systems & Data Collection
- Camera/lens/lighting selection; rig setup & hygiene signage.
- Per-station calibration; ISP tuning (WB/CCM/gamma) when required.
- Data pipelines: retention policy, privacy & consent notes.
Delivery Artifacts
- Pilot report & model cards; acceptance runbook; dashboard starter.
- Deployment scripts & configs with rollback plan.
- Handover & staff training materials.