AI-driven food science intelligence

AI that learns from your batches to predict quality.

Train prediction models on your own production data and get AI interpretation of every analysis.

Specification checks and standard statistics form the trusted foundation the AI builds on. Every model shows its accuracy, and every decision stays with your scientists.

01 / Composition02 / Formulation03 / Quality control04 / Predictive models
Analysis consoleREF-2026.09 · captured September 2026
Protein Bar — Prototype 04Reference data
Moisture
8.4%
Protein
21.8g
Fat
7.2g
Carbohydrates
34.1g
pH
6.12
Water activity
0.48aw
Moisture (%) — batch trendlast 8 batches
ParameterObservedStatus
Protein21.8 gok
Moisture8.4 %ok
Water activity0.48 awok
pH6.12watch
AI Insight

Prototype 04 shows improved protein density and lower moisture relative to the selected baseline. Review texture and sensory-panel results before finalizing the formulation.

Decision-support output — review before acting

Rule-based specification check · requires bench validation

Stability assessmenta_w 0.48 · below 0.60 critical limit
Select a product above to compareFixed reference baseline · not live production data
Live model

A prediction model trained as this page loads

The same ridge-regression engine scientists use in the workspace, trained here on 36 synthetic reference batches and scored by cross-validation.

Model card · water activity predictor

RR-1.0 · trained live in your browser just now

Model type
Ridge regression (L2, λ = 1)
Version
RR-1.0
Target
Water activity (aw)
Inputs
Moisture %, protein %, fat %
Training examples
36
Validation
5-fold cross-validation
CV R²
0.973
CV RMSE
0.0079
CV MAE
0.0073
Mean-baseline RMSE
0.0486

Standardized coefficients: moisture +0.0433 · protein -0.0121 · fat +0.0034

Training data: 36 synthetic reference batches generated from a known formula with added noise, so anyone can reproduce this result. It shows how the method is trained and evaluated; it is not a laboratory validation. In a workspace, the same engine trains on a team's own uploaded batch records.

Observed (x) vs out-of-fold predicted (y) aw. Dashed line = perfect prediction.
How our AI works
System architecture

From raw lab data to auditable decisions.

How data moves through Oqelvia — each stage below is built and running today.

NODE_01 // INGESTLive

Lab data intake

CSV batch files validated row by row, stored privately per workspace with a file fingerprint.

NODE_02 // REFERENCELive

USDA FoodData Central

Live ingredient nutrient lookup, each value traceable to its FDC record.

NODE_03 // COMPUTELive

Analysis engines

Versioned specification checks, formulation math and sensory statistics run server-side.

NODE_04 // GOVERNLive

Audit & reports

Immutable audit trail, role-based access, PDF/CSV reports built from stored records.

Data flow · every stage implemented and running in the current platform

100.0%

Formulation balance

Saves are blocked unless ingredients total exactly 100%, with every version kept.

FDC

USDA nutrient lookup

Ingredient values pulled live from USDA FoodData Central with record links.

p < 0.05

Sensory statistics

Rating and triangle tests, checked against a standard statistics package.

Analytical views(02)

Scientifically readable views of your product data.

Composition comparisons, ingredient contribution, sensory profiles and batch trends — rendered for interpretation rather than decoration.

Composition — current vs baseline

g / 100 g · reference standard

Ingredient composition (%)

% of formulation · reference standard

Sensory profile

Panel mean, 0–9 scale · reference standard

AI Insight

Reducing date syrup by 8% may lower total sugar while maintaining the current formulation structure. Validate texture and sensory acceptance experimentally.

Decision-support output — review before acting

Technical philosophy(03)

Built for evidence-driven food science.

Analytical outputs are only useful when their inputs, models, and limitations are visible. Every insight in the platform is presented for review by qualified food-science, quality, regulatory, or R&D professionals.

Data transparency

Every analysis references the dataset, record count, and parameters it was derived from.

Model transparency

Model type, version, training date, and evaluation metrics are shown alongside outputs.

Human review

AI-generated insights are decision-support outputs, presented for review rather than as conclusions.

Reproducible analysis

Analyses record their inputs and configuration so results can be re-run and compared.

Configurable specifications

Reference ranges are defined by your team, per product and per parameter.

Secure workspaces

Signed-in workspaces are isolated in the database with role-based access, private file storage and audit history. The public evaluation sandbox uses read-only reference data.

Technical evaluation(04)

One reference batch, end to end.

A guided path through the evaluation sandbox. Every step uses the reference high-protein bar batch and runs on the working platform.

  1. STEP 01

    Ingest dataset

    Upload a lab CSV. Rows are validated and bad values listed before anything is saved.

  2. STEP 02

    Run spec check

    Moisture, protein and fat are compared to specification limits on the server.

  3. STEP 03

    Review deviation

    Example: moisture above limit is flagged with a suggested investigation, not a verdict.

  4. STEP 04

    Trace ingredients

    Check ingredient nutrients against USDA FoodData Central record numbers.

  5. STEP 05

    Export & audit

    Download a PDF or CSV report; the action is recorded in the audit history.

Data governance — implemented today
  • Role-based workspaces: owner, admin, scientist, viewer — enforced in the database
  • Immutable audit history for team, formulation and data changes
  • Private dataset storage, scoped per workspace
  • Scoped API keys, stored hashed, rate-limited at 60 requests per minute
  • Verified email, sign-in lockout and 30-minute idle sign-out
  • Reference sandbox kept separate and read-only
Why Oqelvia

Built for food-science workflows

Formulation balance, specification checks and sensory statistics — not generic dashboards.

Scientists stay in control

Outputs are decision support for qualified review, never automatic approval.

Evaluate before you commit data

Explore workflows on reference batches before uploading proprietary formulations.

Get started

Deploy Oqelvia across your R&D and quality teams.

Evaluate composition, specification compliance and sensory panels on pre-configured reference batches — or provision an isolated workspace for your facility's proprietary formulations.