Delectable AI Shared asset library · public
Public asset library

The agentic commerce platform, layer by layer.

Single source of truth for Delectable AI’s product demos, architecture deep‑dives, and technical thought‑leadership content. The platform composes the way a real one does — a foundation of data, knowledge, governance and identity; an intelligence layer of models and pipelines; composable modules; and the agentic experiences shoppers touch. Pick a layer to explore its assets.

▶ Open the Persona Studio — live emulator 🛒 Perfect Cart Comparative Sandbox ⚙︎ Agent Admin Console — reference UX 🗺 Product Roadmap — live tracker 🔑 Portable DID Explorer Sandbox 🛡 OIDC & Web3 Trust Gateway

The platform stack

Top-down: what the world touches sits on progressively deeper substrates. Each layer is independently useful and compounds the value of the layers below it. Click a layer to jump to its assets.

Building diligence? The Evidence tab collects the architecture deep-dives and the Challenge thought-leadership series that explain why the stack is defensible.

Layer 4 · The substrate

Foundation — data, knowledge & trust

Everything above assumes this layer. The canonical Food & Shopper Knowledge Graph and AI-ready catalog data give the platform something true to reason over; governance, identity, integration and multi-tenant config make it safe and deployable per tenant.

Architecture maps

Platform Architecture — interactive module map

The whole stack in one view: experiences → applications → agents → applied intelligence → the Grocery Brain → orchestration → data. Grounded → governed → activated, with Commerce Anywhere (A2A/UCP) and consumer-owned data (Web3/DID). Expandable layers, light/dark, Now/Next/Vision roadmap. The ML detail lives in the Intelligence layer.

platform-architecture.html
Knowledge & data (Information Architecture)

Food & Shopper Knowledge Graph

The HyperGraph — products linked to ingredients, nutrition, flavor, cuisine, dietary tags and recipe primitives. The moat under everything.

delectable-knowledge-graph.html

Knowledge Graph — interactive explorer

Drill from any node — shopper, product, recipe, ingredient, dietary, flavor — into its semantic neighborhood. Inspect edge metadata and see the equivalent Cypher firing against Memgraph.

graph-explorer.html

AI-Ready Catalog — conversational attributes & GEO

How AXP emits structured conversational attributes (MerchantListing JSON-LD, Q&A chunks, variant matrices) so ChatGPT, Perplexity, Gemini & Claude cite retailer SKUs. Up to +40% citation lift (Princeton/GT KDD 2024). Includes live SKU explorer.

ai-ready-catalog.html

Data Foundation — sources to substrate

The data journey: heterogeneous sources (transactions, catalog, recipes, social, nutrition) → ingest → food-science enrichment → the Knowledge Graph, embedding space & feature store the agent reads at runtime. Bridges Foundation → Intelligence.

data-foundation.html
Trust — governance & safety

AI Governance · food safety · audit-ready

The four-pillar substrate that earns trust by construction: hard rules (allergen, dietary, recall), regulatory compliance (FDA · FTC · state, pre-validated at emit), audit trail (every decision logged, traceable in <5s), and model provenance (constrained to the tenant's verified live catalog — never invented).

ai-governance.html

Continuous Compliance, DLP & Evidence

Detailed visual evidence of CCPA/GDPR/CPRA, HIPAA (Zero-Knowledge, clinical shields), AI Governance (hallucination checks, Zero Data Retention proxies), and Infosec audits (SOC 2, CMEK, TLS 1.3, SLAs) with active DLP edge sandbox.

compliance-evidence.html
Integration & partner systems

Integration flows

How retailer data, agent calls, and module surfaces tie together end-to-end.

delectable-integration-flows.html

Instacart integration architecture

Delectable + Instacart integration topology — how a partner fulfillment + catalog network plugs into the agent.

instacart-architecture.html
Identity, config & the multi-tenant backbone

CMS — content, branding & entitlement backbone

The multi-tenant backbone behind every console: per-tenant branding, content, feature flags and entitlement. Resolves through infrastructure/customers/<slug>.json.

cms-overview.html

Identity & Access — multi-tenant RBAC

The trust spine: organizations entitle modules (via_tenant_features), roles grant read/write/admin per module, and every grant is capped by the tenant's entitlement. Interactive role × module permission matrix.

identity-access.html

Centralized Trust Gateway — OIDC & Web3 Proxy

The security traffic cop: unifies enterprise federated SSO (OIDC) and cryptographic Web3 wallets (SIWE) into a uniform, signed JWT header context. Zero database cold starts.

governance-gateway.html

Config, not Code — hierarchical resolution

Every URL, secret, model and feature flag resolves through global → customer → env → module → secret store → process env. Pick a key and watch the precedence ladder resolve and the winning layer glow. Zero hardcoded values.

config-not-code.html

Portable Identity & Compliance — DID Explorer Sandbox

fully-interactive Sandbox proving zero-cold-start storefront personalization, cardiologist prescription compliance, and active cryptographic shredding of envelope-encrypted CAS pointers for immediate GDPR Right-to-Erasure.

did-explorer.html
Layer 3 · The working brain

Intelligence — ML, models & AI pipelines

The data-science layer that turns the foundation into behavior: propensity and household models, the embedding space, the grounding/routing/evaluation pipeline, and the agent architecture that ties them together.

Models & pipelines

Intelligence & ML Architecture — interactive model map NEW

The full intelligence engine in the Platform-chart visual language: model portfolio (graph ML · embeddings · BQML · optimizers · grounded generation), governed runtime (control-plane PEP), data substrate & learning loop. Composed score functions — the LLM is the last-mile narrator. Expandable layers · Now/Next/Vision. Print/PDF companion in doc/.

ai-architecture.html

Perfect Cart Sandbox — Multi-Model Orchestration NEW

Experience how 7+ core models (Bayesian pantry decay, propensity clickstream, GNN householding, IPL optimization, dynamic coupons, last-mile auction) assemble in real time to generate the perfect basket, side-by-side with a legacy search-based cart.

perfect-cart.html

AI Capabilities Deep Dive NEW

8 interactive widgets demonstrating the ML models under the hood: Real-time propensity, GNN householding, orchestrator routing, pantry forecasting, agentic testing, and more.

ai-capabilities-deep-dive.html

Grocery Brain Q&A CMS READY

Interactive marketing-safe answers to the 10 core Grocery Brain questions, with structured content fields ready to migrate into CMS entries.

grocery-brain-qa.html

ML personalization pipeline

Propensity model training, feature store, household GNN, attribute sensitivity — from raw transactions to a per-shopper inference.

ml-personalization-pipeline.html

Agent architecture

How the agent reasons, routes between models, and grounds its answers in the HyperGraph rather than free-associating.

eagle-ai-agent-architecture.html

Embeddings projector

3D PCA walk through the product embedding space — 550 pre-projected points across 9 departments. See how semantically-near SKUs cluster.

embeddings-projector.html
Measurement & observability

Evaluation tracker

Hallucination scorecards, accuracy benchmarks, latency budgets — how the intelligence layer is held accountable over time.

evaluation-tracker.html

Data insights dashboard

Live shopper-behavior analytics on the production data warehouse; exports a 9-tab deck.

data-insights-dashboard.html

Want the defensibility argument behind these models? The Evidence tab carries the deeper reads — the AI pipeline, householding & GNN, and LLM-levers essays.

Layer 2 · Composable products

Module suite

The platform stack, built on the Agentic Commerce Protocol (ACP). Each module is independently entitled and deployable; tenants opt in via the multi-tenant config in infrastructure/customers/<slug>.json.

Protocol & platform

Agentic Commerce Protocol

The contract: A2A capabilities, MCP tools, A2UI events. The model- and channel-portable foundation for every other module.

acp-overview.html

UCP gateway — interactive

Walk through how Google Search calls a grocer endpoint hosted in the grocer's own GCP project. Watch which services light up; track consumption attributed to the customer's billing.

ucp-interactive.html

Unified Commerce Platform — live frame inspector

Three pillars in one frame: Food Intelligence Graph + Shopper Memory + Cart Builder. Switch personas, run a query, and watch the unified context object mutate in real time.

unified-commerce.html
Marketplace, social & operations

Delectable Ads — interactive

Agentic retail media network. Brand-attribute targeting and live ad placement.

ads-interactive.html

Delectable Ads — platform demo

Full RMN platform walkthrough: campaign authoring, attribution, ROAS reporting.

ads-platform-demo.html

Delectable Social — concept

Shoppable discovery pitch: viral trend detection, creator-driven cart, real-time intent signals.

social-interactive.html

Social intelligence pipeline

“TikTok to cohort in seconds” — the pipeline that turns a viral food trend into a merchandisable cohort.

social-intelligence-pipeline.html

Social Operations Admin Console — live

Operator setup console for the social listening fleet. Configure active platform streams, run live Gemini video parser simulations, moderate nutrition claims, and review sponsor match biddings.

social-admin-ux.html

Operator Studio — merch & marketer control surface

The levers a merchandiser or marketer uses to steer the agent. Three tabs — Levers, Test (A/B harness, holdouts, p-values), and Attribution (multi-touch paths, ROAS matrix, incremental lift). Every change auditable.

operator-studio.html
Layer 1 · What the world touches

Agentic experiences

Live retailer demos showing the Delectable AI agent embedded in real-world commerce surfaces — the layer a shopper actually sees.

Composable Social Experience LIVE PRODUCT

The actual shopper-facing storefront with shoppable videos, creator spotlights, trending signals, AI shopping assistant, and a CMS that composes it all per tenant. Opens in a new tab.

delectable-prototypes.web.app/social-experience →

Persona Studio — live emulator HEADLINE

Walk five shopper archetypes across the whole journey — home → category → product → cart — on desktop and mobile, with the agent driving every surface.

persona-studio.html

Split-Stage Comparative Simulator COMPARATIVE

Side-by-side e-commerce face-off. Watch a traditional static store run alongside Delectable's agentic commerce protocol across 4 interactive shopper missions.

split-stage.html

Perfect Cart Comparative Sandbox SANDBOX

A side-by-side comparative playground: Standard Cart vs. Delectable AI Perfect Cart. Toggle between Business Outcome view and Technical Deep-Dive view to see how 7+ orchestrated models build the optimal checkout basket.

perfect-cart.html

Agentic home page

Personalized agentic landing experience with embedded chat assistant.

integrated-home.html

Agentic product page

Product detail view with agent-driven recommendations and substitutes.

integrated-product.html

Agentic shop view

Full agent-driven shopping flow from intent to cart.

integrated-shop.html

Category page

Personalized category browse with propensity-driven ranking.

ge-category.html

Mobile experience

Mobile app frame — the agentic surface in a phone shell, with A2UI cards and image fallback.

mobile-experience.html

Reference UX — built by the Delectable AI UX team

The polished consumer + admin experiences that the platform ships with. Investors comparing the Persona Studio (curated mocks) against the production-grade admin console see both the storytelling and the actual product.

Delectable AI Agent Admin Console — live event timeline, shopper profile panel, enrichment toggles

Agent Admin Console DEV

Live SSE event timeline, waterfall, debug trace. Loyalty ID input, enrichment toggles (promos / sponsored / social / trending), prompt presets. The investor-grade transparency surface.

delectable-ai-stage.web.app
Delectable AI Shopping Assistant — consumer-facing login + chat

Consumer Shopping Assistant AUTH

Production consumer app. Mobile-first, login-gated. Loyalty card sign-in, persona-aware chat, meal plan / shopping list / pantry tabs, recipes browser, profile management.

delectable-ai.web.app
Cross-cutting · Diligence

Architecture & technical evidence

Deep-dives written for engineering and product diligence audiences — the “how it’s actually built” reading behind every layer of the stack.

Interactive architecture

Click through the full system topology: agents, graph, data layer, deploy targets.

architecture-interactive.html

Architecture evidence

The diligence-grade write-up: what's real, what's proven, and where the substrate lives.

architecture-evidence.html

Before / After

Search-era grocery vs. agentic grocery — side-by-side outcomes.

before-after.html

Compliance & Evidence — Trust Registry & DLP

Continuous compliance evidence for HIPAA, CCPA, GDPR, and AI Governance with real-time edge DLP filters, zero-retention audits, and cryptographic proofs.

compliance-evidence.html

The Challenge series

Technical thought leadership: how we build defensible AI for grocery without falling into LLM-wrapper traps. The analyst/exec leave-behind set.

Executive summary

The full thesis in one read.

challenge-executive-summary.html

The AI pipeline

How the agent is grounded, routed, and evaluated end-to-end.

challenge-ai-pipeline.html

Why not just ChatGPT?

The defensibility argument: HyperGraph is the moat.

challenge-vs-chatgpt.html

LLM levers

Where the prompt, the graph, and the model interact.

challenge-llm-levers.html

Householding & GNN

Graph neural networks for household-level inference.

challenge-householding-gnn.html

Algorithmic curation

How rankings stay relevant without becoming homogenous.

challenge-algorithmic-curation.html

Technical deep dive

Architecture for engineering audiences.

challenge-technical-deep-dive.html

Delectable value

The unit-economics story behind the platform.

challenge-delectable-value.html