INLINE ΔE PREDICTION
Spectrophotometric shade prediction against the brand standard, with early drift alerts long before the end-of-batch read.
STAGE 02 // PERCEIVE
Serumon’s perception layer: inline spectrophotometric shade prediction (ΔE against the brand standard) plus high-speed cosmetic-defect vision across compacts, bullets, jars, tubes and bottles. It predicts shade drift early, prescribes the correction back to Emulson, and gates Fillix so only in-spec, defect-free product is filled.
$ shadera watch --sku N-220 SPECTRAL READ 400-700nm 10nm step dE00 vs STD 0.42 hold threshold 0.80 TREND (30 min) +0.09 / h drift detected PREDICT @ EOB 0.71 MARGINAL PRESCRIPTION -> emulson red iron oxide dispersion -0.4 kg yellow iron oxide +0.2 kg confidence 0.91 | awaiting approval DEFECT VISION 3,412 units 2 rejects FILL GATE OPEN
THE PROBLEM
Foundations, lipsticks, powders and creams must hit a brand shade to an imperceptible tolerance, batch after batch. Cosmetics colour is still judged by eye against a standard, so whole batches get held, re-shaded or scrapped over a barely perceptible mismatch.
Surface defects — bubbles, streaks, fill lines, cracked compacts, chipped bullets, cap and label faults — have to be caught inline, but lab inspection is slow, subjective and blind to real-time drift.
The shade matchers who can call it are a handful of people covering dozens of lines.
WHAT A SHADE HOLD COSTS
ILLUSTRATIVE SITE FIGURES [ASPIRATIONAL UNTIL MEASURED].
CAPABILITIES
Spectrophotometric shade prediction against the brand standard, with early drift alerts long before the end-of-batch read.
High-speed surface, fill-level, decoration and cap or label inspection across compacts, bullets, jars, tubes and bottles.
Not just “off shade” — which pigment or dispersion to trim and by how much, fed straight back to Emulson.
A hard gate into Fillix so out-of-spec or defective product never reaches jars, with hold and escalation workflows.
Shade standards and defect images retrievable by similarity (pgvector), with full traceability to batch and lot.
Rare shade-drift and defect variants generated synthetically so the models see failure modes your line has not produced yet.
THE LOOP
Spectral heads and cameras sample the batch and the pack at line speed, normalised for illuminant, geometry and substrate.
Shade ΔE is projected forward to end of batch, and defect classes are scored per unit with confidence.
If drift threatens the hold threshold, Shadera computes the pigment or dispersion correction and hands it to Emulson.
The fill gate opens only for in-spec, defect-free product; everything else routes to hold with its evidence attached.
AI CAPABILITIES
Serumon labels its own maturity. Anything not yet validated in production is marked aspirational, on this site and in the room.
REAL
Fine-tuned regression across pigments and lots, calibrated per substrate and format.
REAL
Multi-format cosmetic-defect and fill-level computer vision running at full line rate on the edge GPU.
PARTLY ASPIRATIONAL
Shade-drift prediction and correction recommendation are real; fully autonomous correction is gated on per-site validation.
REAL
Replicator-generated rare shade and defect variations to cover the long tail without waiting years for real failures.
NVIDIA ALIGNMENT
Line-side inference for spectral models and multi-camera defect vision with deterministic latency.
High-throughput sensor and video pipelines for real-time inspection at line speed.
Quantised vision and spectral models so full-rate inspection fits in the frame budget.
Synthetic generation of rare defect and shade-drift variations for long-tail coverage.
MEASUREMENT
PILOTS ARE CONTRACTED AGAINST A NAMED ROI METRIC AND A BASELINE WE RECORD BEFORE ANYTHING CHANGES.
THE MOAT
Colour instruments and inspection vendors are excellent at producing a number or a reject signal. Neither closes the loop back into the compounding decision that caused the drift.
Shadera accumulates spectra, pigment-lot behaviour, defect images and the corrections that actually fixed them — a paired dataset of perception and successful intervention that no instrument vendor is positioned to build.
DEPLOYMENT SHAPE
ECOSYSTEM
Every agent hands the next one a better decision. Buy one; the others make it sharper.
undefined
undefined
undefined
undefined
SHADERA FAQ
It complements it. Serumon reads from your existing instruments where possible and adds inline sampling where the coverage gap sits. Your instrument stays the reference; Shadera adds prediction and prescription on top.
Yes. Standards are per SKU, per substrate and per illuminant condition, versioned like any other spec artefact. Serumon learns your acceptance behaviour, including where your matchers are tighter than the written tolerance.
They are harder and we say so. Multi-angle measurement is required, and effect-pigment SKUs generally need a longer calibration period before the model is trusted for prescription.
Not necessarily. Many sites keep their existing inspection for pass/fail compliance and use Shadera for prediction, classification and closing the loop back into make.
WEDGE PILOT
Instrument, baseline, shadow, then assist — against a metric you name before we start. If the numbers do not move, you have lost a quarter of telemetry and gained a very detailed picture of your own process.