AI you can trust. Evidence you can trace.
Whatever the data, every decision shows its evidence.
SparkSoft helps manufacturers, labs and commercial teams bring explainable AI into daily decisions across images, tables, signals over time and text — without replacing the workflow you already use.
BACKED BY INNOVATE UK · SOVEREIGN AI PIPELINE PHASE 2 · ONE OF 17 UK COMPANIES
REGION-LEVEL EVIDENCE, FOR AND AGAINST, DRAWN ON EVERY FRAME
…invoices are payable within 45 days of delivery, and the volume rebate applies above 2,000 units per quarter…
EACH RESULT SEALS A DECISION EVIDENCE RECORD · ILLUSTRATIVE
One assurance contract
Explanation looks different for each kind of data. The record does not.
Images
Region-level evidence drawn on the frame — what mattered, where, for and against the call.
Tables
Feature attributions and comparable past cases, ranked by how much each one weighed.
Signals over time
The windows in time that drove the alert, shown against the baseline they broke from.
Documents & text
Highlighted passages and citations behind every extracted answer — checkable at the source.
The product suite
Start with the workflow you already have.
Three products, one assurance contract — adopted as your evidence matures, not all at once. The order is a recommendation, not a requirement: start where the operational need is strongest.
01 · START
SparkSoft Assure
Bring Assure into the workflow you already run. Every important result gains an explanation, an honest confidence level, monitoring and a human review workspace, recorded end to end.
Explore Assure →02 · EXPAND
SparkSoft Pipeline
When you're ready to standardise the wider workflow, Pipeline takes the journey from raw data to a deployed, validated workflow — built from approved components, run where you need it.
Explore Pipeline →03 · AUTOMATE
SparkSoft Automate
Once evidence and human controls are proven, Automate puts trusted results to work — scheduled reports, alerts and approved actions, each with identity, policy and a full history.
Explore Automate →The assurance contract
One record behind every important decision.
Whether the input is an image, a table, a signal over time or a document — every result that matters produces a Decision Evidence Record: what went in, what the AI did, what the evidence was, who decided and what happened next.
- 01Data & context
Input, where the data came from, permissions and purpose.
- 02Model & run
Model, dataset and configuration versions.
- 03Result
Detection, prediction or recommendation, with how confident the system is.
- 04Evidence
Concepts, comparables, citations and tests behind the result.
- 05Human review
Accept, reject, escalate or override — with the reason.
- 06Audit & feedback
Tamper-evident history, monitoring and a controlled learning loop.
Assure on the inspection line
From detection to a decision you can defend.
Detect
The AI flags a possible defect, with the image regions and a confidence score — and is designed to extend to defect types your line has never seen.
Explain
Beside the call sits its case: the strongest evidence for and against, ranked by how much each piece mattered.
Challenge
Your reviewer has the last word — confirm the call, overrule it or send it up the line. Nothing is decided by the model alone.
Trust — or not yet
Only when the same evidence keeps appearing for the same reasons, part after part, does the AI earn a seat in the quality decision.
Industrial inspection
Defect detection with visual explanations, historical defect search, engineer review and audit-ready quality reporting — starting inside the inspection workflow you already run.
Agriculture & bioscience
Image and morphology analysis for laboratory and field, sample prioritisation, specialist validation and case-history reporting.
Commercial decision intelligence
Pricing, risk and anomaly detection on business data and documents, with reasoning from comparable past cases, approval routing and management reporting. Now in build for a global music rights business: a licensing price assistant whose every prediction shows its reasoning.
The programme
Deployable and Scalable Symbolic Capsule Neural Networks for Auditable Zero-Shot Object Detection
Selected by Innovate UK under Sovereign AI Pipeline Phase 2 — one of 17 companies across the UK, backed by a grant of around £770k. In plain terms, the programme funds detection designed to name its evidence and flag defect types it was never trained on. Now identifying manufacturing use cases and design partners with the AMRC — the University of Sheffield's Advanced Manufacturing Research Centre — for a demonstrator that proves technical performance and business value together.
Funded by
Grant record available on request for due diligence.
