Across most industries, quality relies on a human visual check — someone looking and making a call. Inconsistent, unrecorded, impossible to improve at scale. We're building the system that changes that.
The platform we're building encodes human visual judgement into something more measurable, auditable, and scalable. That's the category we're building toward. Right now we're at the exploration stage: prototyping, testing, and looking for the right collaborators to shape what it becomes.
The Builder & The Build
Fintech, risk modelling, data-intensive product design — owning systems end-to-end, from architecture to production. The discipline that runs through all of it: take something complex and subjective, and reduce it to something measurable, auditable, and scalable.
Inconsistent, unrecorded, impossible to improve systematically. The same problem exists across hospitality, agritech, pharma, manufacturing, and logistics — and no one has solved it at scale.
Vision models that can reason about object-level quality now run on commodity edge hardware. Inference costs that were prohibitive at scale two years ago are viable in production environments today. The technology is ready — the category infrastructure hasn't been built yet.
Building early systems that quantify visual standards — not claiming the problem is solved, but demonstrating it's solvable. DishMark is the first local prototype. Five further concepts map EyeMark's cross-industry reach.
Seeking design partners with a real quality problem, technical collaborators who want to work on something genuinely hard, and commercial partners with industry reach. The goal: 2–3 high-quality conversations that turn into real-world deployments.
Swap in a domain-specific model and EyeMark runs in a new industry.
Define
A reference standard is set. What good looks like, captured once.
Capture
Product is captured at the point of inspection, automatically.
Score
A result comes back against the standard — pass, fail, or deviation noted.
Record
Every result is timestamped and stored. Nothing relies on memory.
Continuous visual monitoring at the point of inspection captures the product at the exact moment of handoff.
A detection model trained for each use case isolates the object under inspection and extracts the region of interest — running entirely on-device.
The detected image is validated against a structured definition of the expected quality standard before scoring proceeds.
A large vision model scores the validated image against a defined benchmark — returning a structured quality score and a reasoning trace.
Every score is timestamped and persisted — building a continuous time-series record for each product, batch, or session.
A hosted dashboard surfaces aggregated metrics, trends, and alerts — giving managers a real-time view of quality over time.
Scored results flow automatically into the tools your team already uses — no manual data entry, no separate reports to reconcile.
Quality events are shared with the right people the moment they happen — alerts, summaries, and threshold notifications delivered where they're needed.
Over time, the data reveals what manual checks never could — which suppliers are slipping, which shifts perform best, where quality drifts before it becomes a problem.
The more the system is used, the better it gets. Each inspection makes the next one more accurate — improving automatically without manual intervention.
A restaurant. Evening service. 180 covers. No extra staff. No process change.
Setup
A camera fixed above the kitchen pass, trained on the restaurant's own plating standard — not a generic benchmark.
Score
Every dish is captured at the point of hand-off. EyeMark checks portion, presentation, and consistency against the reference in real time.
Record
Each result is timestamped and logged — pass, fail, or deviation noted. No manual entry. No one has to remember anything.
By end of service: a quality record for every plate served. Trends visible across shifts. Problems findable before they become complaints.
DishMark · Live score
Beef Wellington — Table 12
87
/ 100
Each of these represents a real quality problem currently assessed through manual visual inspection. DishMark is a working local prototype. The others are early concepts — mapped to validate that EyeMark generalises. If you recognise your industry here, that's the signal.
The kitchen pass, made objective.
The first deployment of EyeMark — applied to hospitality. A camera fixed at the kitchen pass scores every dish against a defined presentation standard before service. Consistency, portion, and plating assessed automatically, producing a real-time quality record across every shift.
From paddock to export. Nothing leaves ungraded.
Visual monitoring fixed above the packing house conveyor detects defects, assesses ripeness, and grades produce automatically — replacing inconsistent manual inspection. Every batch exits with a timestamped quality record, giving managers and exporters objective evidence at the point of dispatch.
Reference-standard quality. No fixed hardware.
A mobile-first variant of EyeMark — a single image captured at the point of dispatch scores bloom stage, stem quality, and bunch uniformity against a defined reference standard. Demonstrates that EyeMark operates without fixed infrastructure, opening the engine to distributed and field-based operations.
End-of-line inspection that doesn't miss.
Automated visual inspection at the packaging line detects misaligned labels, damaged seals, and incorrect fill levels before product leaves the facility. Timestamped pass/fail records replace end-of-line spot checks — giving quality managers a continuous, objective audit trail across every SKU and shift.
Contamination, fill, integrity — proven at every unit.
Vision-based inspection of vial fill levels, particulate contamination, and cap integrity — producing a fully auditable quality record at the unit level. Designed for the precision and traceability requirements of pharmaceutical manufacturing, where compliance is non-negotiable and manual inspection cannot scale.
Defects caught on the floor. Not on site.
Structural components inspected before dispatch — surface cracks, dimensional variance, and coating integrity scored against a defined tolerance standard. An objective quality record travels with every part, replacing manual sign-off with consistent, repeatable evidence that holds up under scrutiny.
Not a sales funnel. A genuine search for people who recognise the problem — and want to work on it seriously.
Most quality standards are still
We're exploring how to make them
Companies with a visual quality problem — hospitality groups, food and retail, logistics QA, manufacturing inspection. Any environment where quality is judged visually, by people, and varies because of it.
Computer vision, edge inference, MLOps, infrastructure — especially if you've been frustrated by how poorly lab-grade systems transfer to production environments.
Operators with distribution, industry insiders, or investors who understand what it means to standardise quality across fragmented, real-world environments.
Looking for 2–3 design partners to test this in a real-world environment. Actively speaking to engineers interested in edge inference and vision systems.
If you operate in a space where “good” is still subjective — or you've tried to standardise it and felt the limits of doing so — let's talk.