INTELLIGENCE NEEDS
A BOUNDARY.
The product runs on Hostinger with a Node.js application and persistent SQLite storage outside the public release directory. The AI pipeline below is a development design, not deployed model infrastructure.
WHAT RUNS TODAY.
The public demo is local-only. The signed-in workspace uses server-side sessions, a private SQLite database, account-scoped image access and persisted review history. No model API, background inference or factory connection is involved.
THE PLANNED
AI PIPELINE.
Where GPU computing would fit
Visual-language model inference and evaluation may require GPU acceleration as image resolution and workload increase. We intend to evaluate open models and deployment options against real inspection tasks.
Where cloud services would fit
The current workspace uses private SQLite storage on Hostinger. A future AI service may add object storage, a queue and model workers. Model-provider selection, deployment regions and cost controls have not been finalized.
No NVIDIA or AWS service is claimed as deployed. No program membership, cloud credit, endorsement, partnership or benchmark is implied.
MEASURE BEFORE
MAKING A CLAIM.
Proposed evaluation protocol
Agree a narrow inspection task and ground-truth labeling process. Separate training, development and held-out evaluation data by source to reduce leakage.
Evaluate evidence grounding, false positives, missed visible issues, abstention and reviewer correction. Record model version, prompt, image preprocessing, hardware and data provenance before comparing runs.
We have not published an evaluated model or numeric results. Latency, accuracy and operating-cost claims would require reproducible measurements.