A guided architecture journey · 20–30 minutes

How a layered AI platform turns scattered knowledge into customer value.

This is not a list of projects. It is one architecture — a system of systems — designed so that reusable intelligence compounds from a shared foundation up to the ventures that serve customers. Follow the ten chapters below to understand not just what was built, but why it is shaped this way. Every claim is grounded in the portal's documented evidence.

4 realised layers 7 capabilities 7 repositories 153 graph nodes 98% health
Begin the journey ↓
01
Chapter 1 · The Challenge

Knowledge that doesn't compound

A founder operating several ventures accumulates intelligence constantly — from client projects, research, and operational experiments. The platform's own design records the core problem plainly: without structure, that intelligence is “locked inside documents that are hard to query, easy to forget, and inaccessible to AI assistants.”

Left unaddressed, that problem shows up in six concrete ways — each one documented across this portal:

Fragmented knowledge

Scattered across repositories, docs, and tools; no single connected view.

AI loses context

Assistants can't see the whole system, so each conversation starts cold.

Duplicated capability work

The same competence is rebuilt per project instead of reused.

Knowledge trapped in repos

Reusable intelligence buried inside one venture's codebase.

Architecture scattered

Design decisions spread across many places, hard to reason about.

No long-term memory

Insights die unprocessed; the organisation forgets what it learned.

graph LR
  subgraph Without["Independent AI projects"]
    A[Repo A]:::iso
    B[Repo B]:::iso
    D[Docs]:::iso
    C["AI chat<br/>(context lost)"]:::iso
    K[Knowledge<br/>trapped]:::iso
  end
  Without --> Q["Hard to query · easy to forget · can't compound"]
  classDef iso fill:#eef0f4,stroke:#c9cdda,stroke-dasharray:4 3,color:#333;
02
Chapter 2 · The Vision

An ecosystem where intelligence compounds

The answer is not another app. It is an architecture where competence is built once and reused everywhere — so improving one thing at the base multiplies value across everything above it.

The ecosystem is deliberately organised around six architectural commitments:

  • Reusable capabilities — competence documented and owned independently of any repository.
  • Intelligence products — knowledge published once as versioned, immutable packs.
  • Shared platform services — cross-venture capabilities offered as services, not duplicated.
  • Modular ventures — businesses composed from those reusable pieces.
  • Governed knowledge — every significant decision recorded; documentation is evidence-first.
  • Deterministic processing before AI interpretation — rules decide; AI explains.
graph TD
  F[Shared foundation<br/>reusable capability] --> I[Structured intelligence]
  I --> P[Versioned products]
  P --> S[Platform services]
  S --> V[Modular ventures]
  V --> C[Customer value]
  G["Governance + deterministic core"]:::g -.governs.-> I & P & S & V
  classDef g fill:#eceefe,stroke:#8079ff,color:#333;
03
Chapter 3 · The Platform

A layered architecture

The ecosystem is a stack. Each layer consumes the one below and serves the one above; the shared foundation is consumed laterally by all. Open any layer to see the real systems inside it.

graph TD
  L6["6 · Customer Experiences"]:::v --> OUT((Customers))
  L5["5 · Ventures"]:::v --> L6
  L4b["4 · Platform Services · proposed"]:::p --> L5
  L4["4 · Applications & Agents"]:::r --> L5
  L3["3 · Intelligence Products"]:::r --> L4
  L2["2 · Intelligence Platform"]:::r --> L3
  L1["1 · Shared Skills"]:::ok --> L2
  L1 -. reusable capability .-> L2 & L3 & L4
  L4 -.-> L4b
  classDef ok fill:#e6f4ea,stroke:#1f7a3d,color:#333;
  classDef r fill:#fbf0d3,stroke:#9a6700,color:#333;
  classDef p fill:#eceefe,stroke:#8079ff,stroke-dasharray:4 3,color:#333;
  classDef v fill:#eef0f4,stroke:#c9cdda,color:#333;

Each realised layer, its responsibility, and what it exchanges:

LayerPurposeInputsOutputsConsumers
Shared SkillsReusable AI competenceIntentInvoked capabilityEvery layer
Intelligence PlatformTurn knowledge into structured intelligenceKnowledge, signalsKnowledge assetsProducts
Intelligence ProductsPublish versioned packsApproved intelligenceImmutable packsApps & agents
Applications & AgentsAct on intelligencePacks + evidenceAssessments, outreach, briefsVentures
Platform Services proposedReusable services across venturesVentures
Ventures / CustomerDeliver outcomesApp outputCustomer valueThe market

The layers, live from the graph

04
Chapter 4 · How Intelligence Flows

How knowledge becomes customer value

Follow one thread all the way up. Raw knowledge is progressively refined — extracted, structured, productised — until an application turns it into something a customer receives.

graph LR
  K[Knowledge] --> E[Extraction]
  E --> S[Structuring]
  S --> I[Intelligence]
  I --> P[Products]
  P --> A[Applications]
  A --> O[Customer outcomes]
  O -. feedback · planned .-> I
  classDef d fill:#e6f4ea,stroke:#1f7a3d,color:#333;
  class K,E,S,I,P,A,O d;

Feedback from outcomes is designed to return into the platform as new intelligence — a loop that is planned, not yet built.

05
Chapter 5 · Architectural Patterns

The patterns that recur

Beneath the technologies, the same design decisions appear again and again. These patterns — not any single tool — are what make the ecosystem coherent.

Capabilities over applications

Reusable capabilities are documented and owned independently of the repositories that implement them, so the same capability can serve many ventures.

Where it appearsCapability Registry · every capability page

Reusable products over duplicated logic

Intelligence is published once as versioned, immutable packs that consumers pin — not re-implemented per app.

Where it appearsIntelligence Products (intelproducts) · consumed by leadplatform & outreachagent

Deterministic processing before AI interpretation

Rules decide and score; an LLM only explains the result from structured facts. It never forms the opinion.

Where it appearsWebsite Assessment engine · Commercial Opportunity Profile

Shared platform services

Cross-venture capabilities are meant to live as services, not be trapped inside a single venture repo.

Where it appearsProposed Platform Services layer (ADR 0005)

Evidence-first knowledge

Documentation records what the evidence supports; unknowns are labelled, never invented. Health is computed, not estimated.

Where it appearsThis portal · the deterministic Health engine

Human approval before customer interaction

No knowledge asset becomes canonical, and nothing customer-facing is sent, without a human decision.

Where it appearsKnowledge Intelligence · Outreach (“nothing is ever auto-sent”)
06
Chapter 6 · Key Capabilities

The reusable capabilities

These are the ecosystem's reusable building blocks — documented independently of the repositories that implement them, so each can serve many ventures at once.

Capability

Architecture & Docs Governance

Document the system of systems and govern its architecture: repository digests, capability architecture,

operational Azwaan 3 connections
Capability

Commercial Opportunity & Outreach

Turn a website assessment into a Commercial Opportunity Profile (COP) — a structured decision on whether a

operational Azwaan 7 connections
Capability

Intelligence Productization

Package the platform's intelligence into versioned, immutable, machine-consumable packs that downstream

operational Azwaan 10 connections
Capability

Knowledge Intelligence

Turn a multi-venture founder's scattered knowledge into a queryable, AI-accessible, scoped asset library:

early-development Azwaan 8 connections
Capability

Reproducible Evaluation

Provide an immutable, fully-provenanced experiment & benchmark harness so engines (assessment, migration,

early-development Azwaan 5 connections
Capability

Reusable AI Skills

Package domain expertise into modular, invocable Agent Skills so competence is built once and reused

operational Azwaan 13 connections
Capability

Website Assessment Platform

Assess a service business's website — crawl it, score it against an industry intelligence pack, and

mixed Azwaan 14 connections
07
Chapter 7 · Real Platform Journeys

Following three real journeys

The architecture is easiest to grasp by watching real capabilities travel through the layers. Here are three — each crossing several layers on its way to value.

Website Assessment — evidence into a sales conversation

graph LR
  KA["Intelligence Platform<br/>industry knowledge"] --> PK["Products<br/>assessment pack"]
  PK --> ENG["Applications<br/>assessment engine (crawl → score)"]
  ENG --> COP["Applications<br/>Commercial Opportunity Profile"]
  COP --> OUT["Venture<br/>Inexis Digital outreach"]
  classDef d fill:#e6f4ea,stroke:#1f7a3d,color:#333; class KA,PK,ENG,COP,OUT d;

Industry knowledge becomes a versioned pack; the engine applies it to a real website and scores it deterministically; the result becomes a structured opportunity profile; a human approves before any outreach. Rules decide, AI explains, a human sends.

Knowledge Intelligence — capture into reusable assets

graph LR
  SRC["Knowledge estate"] --> CAP["Capture"]
  CAP --> REV{"Human review"}
  REV -->|approve| ASSET["Knowledge assets<br/>scoped · maturity-gated"]
  ASSET --> RET["Retrieval / products"]
  classDef d fill:#e6f4ea,stroke:#1f7a3d,color:#333; class SRC,CAP,ASSET,RET d;

Proposal Intelligence — knowledge into a client-ready draft

graph LR
  PACK["Products<br/>proposal pack"] --> SKILL["Shared Skills<br/>proposal-generator"]
  SKILL --> APP["Applications<br/>leadplatform / outreach"]
  APP --> DRAFT["Draft proposal<br/>(human reviews)"]
  classDef d fill:#e6f4ea,stroke:#1f7a3d,color:#333; class PACK,SKILL,APP,DRAFT d;
08
Chapter 8 · The Ventures

Ventures as compositions

The founder operates several venture areas — Inexis Digital, Inexis Consulting, Inbound Lanka, City Retreats, and cross-cutting work. Crucially, they are not independent software products: each is a composition of the same reusable capabilities.

graph TD
  CAP1["Reusable AI Skills"]:::c --> V1["Inexis Digital"]
  CAP2["Website Assessment"]:::c --> V1
  CAP3["Intelligence Productization"]:::c --> V1
  CAP1 --> V2["Inexis Consulting"]
  CAP3 --> V2
  CAP1 --> V3["Inbound Lanka"]
  CAP4["Knowledge Intelligence"]:::c --> V1 & V2 & V3
  classDef c fill:#e6f4ea,stroke:#1f7a3d,color:#333;

Because capabilities are shared, a new venture is closer to a composition than a fresh build — which is exactly why multiple ventures can exist without duplicating architecture. The systems that realise this today:

SystemShared Skills

Shared Skills Layer

The Shared Skills layer is the foundational capability layer of the ecosystem: a single curated

Composed of
09
Chapter 9 · Technology Choices

Technology in service of architecture

Technologies are chosen for the architectural job they do, not for their own sake. Grouped by responsibility, the choices tell a consistent story: static-first, deterministic-first, contract-driven.

AI runtime & agents

The reasoning layer — kept as an explainer over deterministic logic, not the decision-maker.
Anthropic Claude (Agent Skills, Agent SDK, Claude Code)CodexOpenClaw (self-hosted, planned)

Knowledge & data

Where structured intelligence and application state live.
Supabase / PostgreSQL (knowledge store, system of record)Cloudflare D1 + R2 (application data & media)

Edge & delivery

Static-first, globally distributed, cheap to run.
Cloudflare Pages / WorkersAstro static site generationCaddy · Docker (the evaluation lab)

Applications & crawling

The venture-facing apps and the evidence they gather.
TanStack Start / Next.js / React 19Playwright (website crawling)Node ESM pipelines

Distribution & contracts

How intelligence moves between systems as stable contracts.
Versioned intelligence packs (git submodules)Frozen JSON contracts (COP, assessment)
10
Chapter 10 · The Future

Where this is going

The graph and the deterministic Health engine are the foundation for more. What follows is labelled honestly — implemented, in progress, or planned — and no planned capability is presented as built.

Implemented
  • The Portfolio Graph — one runtime model of the ecosystem.
  • The deterministic Portfolio Health engine.
  • Reusable AI Skills; published intelligence products.
  • The Website Assessment engine (core flow).
In progress
  • Knowledge Intelligence (design + foundation built; capture loop pending).
  • The evaluation lab's core run/eval loop.
  • Live cross-repo assessment (mock today).
Planned
  • A Platform Services layer (extract reusable services — ADR 0005).
  • Portfolio Intelligence — drift, staleness & health interpretation.
  • Hermes — a one-way learning loop over outcomes.
  • The self-documenting orchestrator.

Future Portfolio Intelligence and Hermes will interpret the deterministic outputs you have just toured — not replace them. That is the whole point of building the model first.

That's the architecture — one coherent system, not a collection of projects.