Products
Three products. One assurance contract.
Adopt them in the order your evidence matures: bring SparkSoft Assure into the workflow you already run, expand into SparkSoft Pipeline to standardise it and add SparkSoft Automate once trusted evidence and human controls are in place. Every product writes to the same Decision Evidence Record.
SparkSoft Assure — start here
Beyond the result: see what drives it.
Assure joins the workflow you already run — around your AI or ours — and makes its results explainable, monitored and reviewable without replacing anything. It speaks the language of your data.
Model-neutral by design: keep YOLO, RF-DETR, LightGBM, XGBoost, foundation models or your own approved systems. Assure adds the assurance layer without asking you to displace what already works.
Explanations with honest confidence
Evidence suited to the data, a confidence level you can rely on and a warning when an input looks unlike anything the AI was trained on — beside every important result, so a decision reads as evidence, not a score.
Historical evidence search
A new detection is linked to similar past cases, inspection records and final acceptance decisions, so every explanation connects to real precedent.
Human review workspace
Authorised reviewers accept, reject, escalate or override with a reason, request second opinions and compare against history. Confirmed outcomes feed a controlled learning loop.
Monitoring that catches change
Assure watches how inputs, results and explanations shift over time — what specialists call drift — plus how often reviewers disagree, so a change can be caught before it becomes a quality incident.
Audit-ready records
Every important decision produces a Decision Evidence Record you can query, export and put in front of a customer, a regulator or your own management.
Business-value dashboard
Review time, false-positive cost, escaped defects, acceptance and override rates — the outcomes your management actually measures.
Why it matters
Three things change when the reasoning is visible.
See why the AI made each call
Assure traces each detection back to what drove it inside the AI and describes it in terms an inspector would use. Accuracy tells you how often the AI is right — this tells you why.
Catch the errors that pass every check
Once the reasoning is visible, the confident-but-wrong calls stand out: detections whose evidence sits on background, glare, a fixture edge or a printed label. Those look correct in testing — and fail on the line.
Know what to improve next
Every decision also names the evidence arguing against it, so you know which conditions would make the AI miss — and can point new training data, extra examples and threshold tuning at a measured weakness instead of a guess.
SparkSoft Pipeline — expand
From raw data to a workflow you can operate.
Pipeline combines data connection, preparation, workflow design, model validation and deployment into one journey. Check and prepare your data, assemble approved components, train or attach models, compare alternative set-ups side by side, validate the workflow — then run it the way your environment requires.
Data readiness, assessed first
Gaps, duplicates, inconsistent formats, image quality and how well your images are tagged, sensitive information and data rights — a clear assessment of whether your data is ready, with an action plan, before you commit to a larger programme.
Approved, reusable components
Workflows are assembled from governed components and templates rather than bespoke engineering, so the next use case costs less than the last.
Deployment where the data lives
On a schedule, on demand through an API or triggered automatically by your systems — running in your private cloud, on your own servers or directly on equipment at the plant. The same validated workflow, wherever your data must stay.
SparkSoft Automate — when trust is earned
Automation that stays inside the guardrails.
Automate turns trusted results into scheduled reports, alerts and approved operational actions, using controlled agent tooling built on MCP (the Model Context Protocol), an open standard for how AI agents connect to tools and data. It is deliberately the last step of the journey — a credible route to automation, not a request to trust autonomous agents on day one.
Read-only by default
Agents submit data, run approved workflows, retrieve evidence and generate reports. Nothing material happens without a policy check — and, where it matters, explicit human approval.
Identity and scope for every agent
Each agent has an owner, a declared purpose, scoped tools, data boundaries, execution limits and an emergency disable.
A full action history
Every agent action lands on the same Decision Evidence Record as everything else: who acted, on whose authority, with what evidence and what happened.
Assure leads the suite today. Pipeline and Automate follow the same assurance contract as your adoption and controls mature — talk to us about the right sequence for your workflow.
Start where the need is strongest.
Bring a model, a workflow or just the decision you need to defend — if the platform doesn't fit, we'll say so.