The platform
Data becomes a graph. The graph becomes a judgement.
Three layers of intellectual property, twenty-seven agents that argue before anything reaches a board, four domain models we own, and seven platform layers you can adopt at any level.
Three layers of intellectual property.
Each layer is defensible on its own. Together they turn raw disclosure into a decision the board can price.
The data
DAK Framework
Data, architecture and knowledge: entity resolution, ownership graphs and source lineage across 10,500+ entities in 55 African states, with listed issuers counted directly from exchange registers.
The method
PSIDA™
A seven-dimensional intelligence lens carrying 47+ KPIs per entity, each confidence-banded and Monte Carlo validated at 10,000+ iterations.
The outcome
Share-Price Engineering™
Value-driver mapping and investor-narrative decomposition that links a management decision to the drivers of enterprise value, typically two to three quarters ahead of a re-rating.
The PSIDA™ lens.
Seven dimensions, 47+ KPIs per entity, each confidence-banded. The seventh dimension, regulatory and risk, is where most competitive surprises actually start.
- Positioning
- Share, pricing power, and where the pool actually sits
- Strategy
- Stated intent versus capital actually deployed
- Innovation
- Product velocity, partnerships, build-or-buy posture
- Digital
- Channel economics, platform maturity, technology spend
- Assets
- Balance sheet, network, licence and franchise value
- Sentiment
- Analyst, funder, press and social narrative
- Regulatory & Risk
- Supervisory stance, licence exposure, enforcement flow
Twenty-seven agents that argue first.
Agents pass structured context across a governed bus and escalate to a human analyst when consensus cannot be reached at the required confidence. That escalation rule is what makes the system deployable inside a regulated institution.
Tier I, 9 agents
Ingestion & enrichment
Acquire, resolve, normalise and lineage-stamp every source.
Tier II, 12 agents
Analysis & reasoning
Score, simulate, cross-examine and contest each conclusion.
Tier III, 6 agents
Synthesis & delivery
Compose, rank, attribute and render for the board.
Output gate
Confidence gate
Publish with a confidence band, or escalate to a human analyst.
Typed context
Every hand-off is a schema-validated object carrying claim, source, timestamp, method and confidence.
Contest before consensus
Tier II agents are mandated to disagree. A conclusion that survives challenge carries more weight.
Least-privilege context
An agent sees only the slice its task requires, which contains the blast radius of any failure.
Sub-second kill switch
Anomalous behaviour isolates and stops the agent in under a second; the run resumes from a checkpoint.
Four domain models we own, and no single vendor.
A task-aware router sends each request down two lanes and reconciles them. When the frontier moves, the router is re-pointed rather than the business re-platformed.
Request
From an agent or an executive
Task-aware router
Sends each request down both lanes
Frontier lane
6+ models for breadth, language and open-world reasoning
AO-series lane
4 models for domain precision on corpora we hold
Reconciliation
Agreement publishes with a confidence band; disagreement escalates
AO-Finance-7B
The quantitative core
Ratio construction, valuation mechanics, revenue-pool decomposition and Share-Price Engineering™ mathematics, calibrated against filed accounts.
AO-Competitive-3B
The move predictor
Competitor behaviour and positioning inference across large entity universes, trained on how institutions have actually behaved, market by market.
AO-Regulatory-5B
The jurisdiction reader
Multi-market supervisory, prudential and AML/CFT analysis under grounded-answer guardrails. It refuses rather than infers where the instrument is silent.
AO-Executive-2B
The board register
Compresses analysis to executive length while keeping the evidence chain. The smallest model, and the one boards actually read.
Seven layers. Adopt at any of them.
You can start at the intelligence-product layer without rebuilding the foundation beneath it. Provenance, audit log and evidence chain travel with every answer, at every layer.
- L7
Experience & executive interface
Role-specific dashboards, board pre-reads, committee briefings, delivered into the channels executives already use
- L6
Intelligence products
The Platform, Revenue Pools, Regulatory Intelligence, Investor & Funder Sentiment, Executive Strategy Engineer™
- L5
Agentic & reasoning layer
27 agents, PSIDA™ scoring, Monte Carlo at 10,000+ iterations, peer cross-validation, a human escalation gate
- L4
Model layer
6+ frontier models and 4 proprietary AO-series models behind task-aware routing, with no single-vendor dependency
- L3
Data & enrichment fabric
150+ sources: market data, regulator and supervisor portals, regional feeds and first-party enterprise data
- L2
Security, identity & governance
Enterprise SSO, role-based access, data residency, immutable audit log, an IP firewall between platform and client estate
- L1
Infrastructure & compute
Sovereign-region hosting, isolated tenancy per client, self-hosted model serving, containerised deployment
Every number carries a passport.
No figure enters an AlphaOmega output without a provenance class, a source, a timestamp and a confidence band. Where a figure is not published, the blank is printed and marked, never estimated and never shown as zero.
Counted
Enumerated directly from a primary register: exchange listings, licence registers, filed returns. The strongest class.
Proven
Stated in an audited or officially filed document, cited to page and date. A client figure is cited as theirs.
Derived
AlphaOmega arithmetic on counted or proven inputs, with the method published alongside the result.
Modelled
Simulation output, confidence-banded and labelled, shown at lighter weight than evidence.
Where a decision surface does not exist, we engineer one.
Two to six weeks to a first working artefact, with acceptance criteria agreed upfront and signed off at the end of every sprint.
Data
Pipelines into your estate
Connections into core systems, transaction platforms, partner interfaces and warehouses, governed and monitored like production infrastructure.
Agents
Custom agent development
New agents built to a named task and added to the governed bus with their own guardrails, thresholds and escalation rules.
Models
Fine-tuning on your corpus
Domain models trained on your own documents, policies and decision history, deployed inside your tenancy.
Interfaces
The surface people use
Role-specific dashboards, board packs, internal tools, and delivery through the channels your executives already use.
Assurance
Evaluation & testing
Evaluation suites and regression tests before any version change reaches production, with accuracy reported for risk and audit.
Deployment
Inside your perimeter
Containerised deployment into your own cloud subscription and identity estate where sovereignty or licence conditions require it.
Bring us one real question your current systems cannot answer.
We scope it as a contained sprint, agree the acceptance criteria in writing, and return with a working intelligence artefact.