Skip to main content

The Architecture of Scale

Framework, product, proof. Built one person at a time.

Most AI strategy decks describe what could be built. This one describes what was. I developed the Three-Pillar Framework while engineering £15M in revenue growth at a national home services business, then applied it inside an established ITSM and HRCM software company to ship a production AI platform, Solvyr®, that ranked first of 47 businesses in the Group's AI Maturity Assessment.

What follows is the framework, what I built with it, and what it delivered.

#1 of 47
Group AI Maturity Assessment
32 points clear of second
15 production tools
shipped in 12 months
ITSM, HRCM, sales, finance, field service
Microsoft co-presented
Quadrants 2026, Copenhagen
Data moat reference architecture
01

Revenue & Growth Intelligence

Engineering revenue predictability through conversation intelligence.

Revenue and Growth Intelligence data flowLive call audio and behavioural data ingested through speech-to-text and LLM reasoning, augmented with risk modelling, and surfaced as forward-looking revenue intelligence with ninety day visibility.INGESTIONAutomatic CaptureSales CallsSupport TicketsEmail ThreadsPROCESSINGAI-Driven NLUSpeech-to-TextLLM ReasoningSentiment ScoringAUGMENTATIONContext EnrichmentBehavioural DataUsage MetricsRisk ModellingOUTPUTRevenue Observability DashboardAt-Risk Deal AlertsChurn Early WarningExpansion SignalsDeal Health ScoreSales Coaching TriggersIntervention PlaybooksCLV OptimisationRevenue ProtectionLive Call Intel Streaming • 90-Day Forward Visibility

Focus:

Engineering revenue predictability through conversation intelligence.

The Problem: Most organisations have a 90% data blind spot. Customer truth lives in unstructured calls, support tickets, and email threads. Leadership runs on gut feel because the signals never reach the dashboard.

The Strategy: Capture every digital touchpoint. Process voice and text through transcript intelligence and sentiment mapping. Score deal health, surface buying objections and renewal risk 90 days before they hit the P&L.

Commercial Result: Move from reactive support to proactive revenue protection. Lift Customer Lifetime Value by intervening on unspoken dissatisfaction before it becomes a churn decision.

Proof in Solvyr®Built as Live Call Intel Streaming. Real-time sentiment with SPIN/MEDDPICC coaching prompts, surfacing open questions and actions during live sales calls. Demonstrated at Quadrants 2026.

02

Governance & Trust

Grounding AI in company truth for compliance and credibility.

RAG architecture for citation-verified retrievalKnowledge base feeding a vector database vault using Azure AI Search with hybrid retrieval, HyDE expansion and Cohere reranking, into an LLM layer that produces citation-verified outputs.KNOWLEDGE BASEPDFs, DocsPolicies & ProceduresLIVE DATACRM, ITSMReal-time FeedsVECTOR DBAzure AI SearchHybrid RetrievalThe VaultLLM LAYEROrchestrationPrompt EngineeringContext AssemblyResponse GenerationOUTPUTCited, Verified ResponsesHyDE · RerankActive Intelligence Layer • 100% Auditable • Citation-Verified

Focus:

Grounding AI in company truth for compliance and credibility.

The Problem: Enterprise AI fails when it lacks a verifiable foundation. Hallucinations and black-box outputs are unacceptable in compliance-heavy industries. Manual tender responses and governance documentation drain hundreds of executive hours every quarter.

The Strategy: Architect a verified knowledge layer using Retrieval-Augmented Generation. Static knowledge bases and live data feed a high-performance vector database. Every response is anchored in cited internal truth before reaching the LLM.

Commercial Result: Week-long manual processes collapse into minutes. The output is 100% auditable, defensible, and scales bid capacity without growing headcount.

Proof in Solvyr®Built as Solvyr® Vault. Governance and compliance documentation engine supporting ISO 27001:2022 certification, with HyDE and Cohere Rerank in the retrieval pipeline. Tender response time reduced by 80%.

03

Operational Excellence

High-velocity AI delivery without the rewrite risk.

Decoupled AI architecture with zero rewrite riskModular AI service layer connected to a legacy core through a secure API bridge, allowing modern AI capabilities to ship alongside the existing system without rewriting it.AI SERVICE LAYERFastAPI • React/Next.js • Modern DevExHigh-Velocity Innovation ZoneAPI BRIDGESecure • Typed • Rate-Limited • CachedLEGACY COREJava Monolith • Decades of Business LogicRequestsResponsesQueriesData10x VELOCITYZERO REWRITE RISK

Focus:

High-velocity AI delivery without the rewrite risk.

The Problem: The most dangerous words in enterprise IT are "let's rewrite it from scratch." Legacy technical debt acts as an innovation tax, but traditional modernisation programmes are too slow, too expensive, and too often fail to deliver commercial value.

The Strategy: Decoupled architecture. Wrap the legacy core in a secure, typed, rate-limited API bridge. Build the modern AI service layer on FastAPI and Next.js, where features ship in weeks without touching the battle-tested backend.

Commercial Result: 10x release cadence. Legacy liability becomes high-speed innovation engine. Engineering risk goes down, not up.

Proof in Solvyr®This is the architectural pattern behind the entire Solvyr® platform. 15 production tools in 12 months, on a 25-year-old Java monolith, with zero rewrite risk.

ACT 02

Solvyr®. The Framework, Productised.

I joined an established ITSM and HRCM software company in October 2024. Twelve months later, Solvyr® was live: a production AI engine spanning the full ITSM and HRCM service lifecycle, plus adjacent tools for sales, governance, finance and field service. One operator, the right tools, a clear framework. This is what got built.

Service Desk Core

04

Auto Triage

99.8% categorisation accuracy at scale

Smart Resolution

Intelligent assistance surfaced at the point of need

Knowledge Creator

Automated synthesis from resolution data

Solvyr® Studio

Turns monolithic documents into a searchable knowledge centre with semantic search and natural language access

Feedback & Communications

03

Survey Sentiment

Real-time feedback analysis

Live Call Intel Streaming

Instant sentiment feedback with open questions and actions

Comms Boost

AI-enhanced customer communications

People & Culture

02

Solvyr® Shield

HR case management grounded in current UK employment law, with a Teams chatbot enabling employees to query the handbook conversationally

Hype

Employee recognition tool

Governance & Compliance

02

Solvyr® Vault

Governance and compliance documentation engine, supporting ISO 27001:2022 certification

Quality Audit

Deep reasoning engine for Gas Safe regulated businesses

Commercial & Operational

04

ROI Calculator & Digital Deal Deck

Quantifying value and packaging the commercial story

PPC Capacity Bridge

Connects Google Ads bidding to live field service capacity for tuned performance

Bank Reconciliation Engine

Automated transaction matching for finance ops

Retention Intelligence

Customer churn prediction and intervention model

The Shared Foundation

Every module draws from and contributes to a single Azure-native data layer. The moat compounds with every interaction.

Solvyr platform architecture across application, retrieval and Azure foundation layersSolvyr application modules sit on a retrieval layer using HyDE expansion, Cohere reranking and hybrid search, supported by an Azure foundation of Azure AI Foundry, Azure AI Search and PostgreSQL with pgvector, connected to a legacy Java monolith through a secure API bridge.APPLICATION LAYERFastAPI services • Next.js front ends • Teams bot integrationsAuto TriageSmart ResolutionSolvyr® StudioLive Call IntelSolvyr® ShieldSolvyr® VaultQuality Audit...and eight more across feedback, HR, governance, sales, finance and field serviceRETRIEVAL LAYERThe query pipeline. Anchored, ranked, hybrid.HyDEHypothetical document expansionCohere Rerank 4Top-of-stack retrieval qualityHybrid SearchVector + full-text fusionFOUNDATION LAYER · AZUREThe shared data moat. Every module reads from and writes to the same substrate.Azure AI FoundryManaged model deploymentClaude · GPT-class reasoningAzure AI SearchVector store. The data moat.Where the moat operationally livesPostgreSQL + pgvectorRelational + vector at scaleOperational state · embeddingsLEGACY BRIDGESecure, typed, rate-limited API connecting to a 25-year-old Java monolith. Zero rewrite risk.Secure API BridgeJava Monolith (legacy core)Battle-tested logic, untouched. Wrapped, not rewritten.
ACT 03

Validated, Not Self-Reported

Internal claims are easy. External validation is harder. Here is what others have said about the work.

AI Maturity Assessment

#1 of 47 Group Businesses

Solvyr® ranked first of 47 Group businesses in the 2026 AI Maturity Assessment, scoring 82%, 32 points clear of second place. The assessment was conducted by the Group’s AI leadership and benchmarked across portfolio.

Commercial Deployment

First Paying Customer Live

Solvyr® went live with its first commercial deployment in early 2026, with additional customers in active onboarding. The product has moved from internal tool to revenue-generating SaaS.

Solvyr® has transformed our Service Desk operation.

IT Manager, launch customer
Group Quadrants 2026

Portfolio Recognition in Copenhagen

I took the Solvyr® story to the Group Quadrants in Copenhagen in April 2026, leading an expert session on AI and Automation That Pays Back, and co-presenting two sessions with Microsoft on the data moat and how Azure AI Search operationalises it. The Group’s portfolio leadership was in the room.

Microsoft Co-presentation

Reference Architecture

The Solvyr® architecture, Azure AI plus Azure AI Search as the operational layer where the data moat lives, was used as the reference pattern in the Microsoft co-presentations. Not a customer testimonial. A reference pattern other portfolio companies were asked to study.

From Volaris Group Quadrants 2026, Copenhagen

Huge thanks to Christopher Carswell [...] for sharing his entrepreneurial wisdom. His passion for building with AI is infectious and perfectly aligns with why I love the Volaris Group Quadrants, it’s innovation with a roadmap. Feeling inspired and incredibly lucky to learn from the best.

Gary Wong, on LinkedIn after Quadrants 2026

The Toolkit

Pragmatic choices for production AI.

🐍
Backend

Python / FastAPI

High-performance async APIs

⚛️
Frontend

Next.js / React

Production front-end framework

🐘
Database

PostgreSQL + pgvector

Vector search at scale

☁️
AI Platform

Azure AI Foundry

Managed model deployment

🔎
Vector Store

Azure AI Search

Where the data moat lives

🎯
Reranking

Cohere Rerank 4

Retrieval quality at the top of the stack

Query Expansion

HyDE

Hypothetical document embeddings

🧠
Reasoning

Anthropic Claude

Primary model for production workloads

🔴
Cache & Queue

Redis

Real-time performance layer

Dev Environment

Cursor

AI-assisted development