Checklist

Edge & On-Device AI Evaluation Checklist

Interactive checklist and evaluation worksheet to determine whether and how to deploy AI models on edge and mobile devices. Covers latency, throughput, device constraints, model size, compression strategies, privacy and data minimization, offline behavior, update and rollback plans, monitoring, and a simple feasibility score to guide go/no-go decisions.

Interactive Tool

Edge & On-Device AI Evaluation

This interactive checklist helps teams assess whether on-device AI is the right choice for a project and documents the practical tradeoffs. Use it to record device constraints, latency and throughput targets, model size limits, privacy considerations, update strategy, monitoring needs, and an overall feasibility rating. Save each evaluation to build organizational memory and compare approaches across projects.

Short name to identify this evaluation (e.g., 'Camera-based Defect Detection v1').
Date of the evaluation (YYYY-MM-DD).
Who completed this assessment and which team they represent.
List specific devices, OS versions, and classes (e.g., 'Android phones w/ARM A53, Edge gateway with NPU, Raspberry Pi 4').
Approximate number of cores available for inference on target devices.
Memory budget for model + runtime + app (in MB).
Answer 'yes' if target devices have an accelerator (NPU, DSP) that the model can leverage.
Maximum acceptable inference latency for the feature (end-to-end), in milliseconds.
Expected concurrent inference rate per device or for the device class.
Storage budget for the model on the device (including possible compressed format).
Select formats you can deploy (pick all that apply).
Techniques that might be applied to meet size / latency constraints.
How strict are privacy requirements for the data and model outputs?
If yes, document retention, minimization, and encryption plans in the notes section.
List how you will reduce privacy exposure (e.g., local-only processing, ephemeral buffers, hashed identifiers).
Describe how the feature should behave without network access and how queued/failed items are handled.
How will you deliver model updates and patches to devices? Consider OTA, staged rollouts, and signed packages.
Which metrics will you collect (inference latency, error rate, model drift indicators), and how will you respect privacy when collecting telemetry?
How will you roll back a problematic model or patch? Include communication and safety measures.
List any legal, medical, safety, or industry regulations that affect on-device deployment.
Rough effort to implement and support the on-device solution (including ops and monitoring).
Rate overall feasibility from 1 (low) to 5 (high) given constraints and requirements.
1.0 10.0
A concise go/hold/stop recommendation based on the evaluation.
Concrete next steps (e.g., prototype quantized model, setup OTA test channel, implement telemetry), owners, and due dates.
Any other observations, risks, or context.
You can explore this tool now. Sign in or create an account to save your responses and return to them later.
Sign In to Save Create an Account
Make this tool part of your work

Save a personal copy, bring it to your team, or tailor the questions and workflow to fit what you are hungry to improve.

Member customization and team collaboration are coming soon.
Discuss Ask the Librarian

Discussion

Comments and conversation will live here.