Platform  /  Architecture

Chart the path ahead

ALPHA gives sophisticated investors and financial institutions a reasoning architecture — risk models and analytics, optimizers, portfolio implementation, semantic data infrastructure, and alignment — that work together to help them reason further than they otherwise could.

Every institution has an inference horizon: the frontier of its reasoning, set by the data and tools within reach. ALPHA expands yours.

What this is

Metrics describe the past.Operational AI optimizes the present.Decision science models the future.

Optimizing the present based on a description of the past is like steering a car by its speedometer, fuel gauge, and tachometer. The instruments are accurate. They are not navigation.

Real navigation requires looking ahead — and the future, by definition, has no map. The firm has to construct one through inference. Not guesswork. Not naive extrapolation. Not consensus. The work is to push the inference horizon outward, to reason further than the firm could otherwise reach — further than your competition.

Operational AI is not decision science.

Operational AI platforms model entities and orchestrate workflows. Incumbent risk platforms give you one enclosed system to move into. ALPHA is neither. It models economic systems under uncertainty — and it is open: the engines embed into the systems you already run, ingest from any source, and harmonize every output back to your own data model. The architecture below is the structure of that difference — three independent capabilities, wired into one continuous reasoning loop.

Plate I The reasoning architecture

How we bring science, information and alignment together to help make better decisions.

ALPHA analyzes billions of data points each day — describing tens of billions of dollars of investment transactions, millions of investors, thousands of investment managers, and nearly every security, currency, and economy on earth.

Information data, ontologies, evidence. Alignment views, beliefs, needs. Decision Science risk models & analytics. One data point of billions
Plate II Component register

The five components, in order.

The animation traces one data point's path. The architecture is what it passes through. Each component is a complete piece of the firm's reasoning.

01

Inbound

Ingestion OpenFinance lineage

The system accepts data from any source the firm uses, plus sources the platform has not seen before. Type inference and lineage attach at the door. 400+ direct connections to banks, asset managers, custodians, and common industry software, plus 1,500+ more reachable through aggregation partners. A new, fully custom source — never seen before — can go from spec to developed, tested, and live in production at 99.9997% accuracy in under two weeks. AI-enhanced integration tooling, 15+ years in production.

400+ direct connections live · 1,500+ via aggregation partners · new sources spec-to-live in < 2 wk

02

Information

PAIX OpenFinance lineage

The semantic data layer. Over $4T in AUM moves through PAIX, typed against an economic model rather than a database schema — positions, securities, transactions, organizations, investors, markets as interconnected entities. The context for reasoning. Reconciled and QA'd against source-of-truth records. Expressive, built to extend — programmatically and by intervention — in production in some form and growing for three decades.

99.9997% post-reconciliation accuracy · batch · event-driven · streaming (dev)

03

Alignment

View · Narrative Market76 lineage

The institution's logic, applied twice. View applies the firm's preferences, constraints, and trade-offs on the way in. Narrative composes the result on the way out — decision, reasoning, justification, in the firm's own terms. Developed in partnership with BlackRock, Man Group, and AIG, from analysis of millions of client relationships across 14,000+ advisors and registered representatives at over 200 firms.

Patents granted and pending (USPTO) · empirical foundation: 14k advisors · 200+ firms

04

Decision Science

Everything Everywhere · Open Optimizer Northfield lineage

Forty years of factor decomposition, scenario inference, and regime detection — developed continuously since 1985, with over 2,000 published papers across the discipline. Risk models that cover everything everywhere, including the flagship Everything Everywhere model, which addresses public, private, and derivative instruments under one model. Open Optimizer composes decisions that respect the firm's view, constraints, and tax position simultaneously.

40+ years in production · 2,000+ papers · global multi-asset coverage in one model

05

Delivery

PAIX OpenFinance lineage

Decisions arrive in the firm's own data model, not the platform's. Automatic harmonization across systems, security masters, entities, semantics — everything. Everything and everyone works together seamlessly and grows together.

API · relational interfaces · file delivery · custom integrations in custom schema

Plate III Reference

How the system is built, and how it runs.

A reference for architects, risk leads, and operators. Four registers: principles, roles, extensibility, operations.

I

Principles

Seven theses
01

Semantic typingnot schema mapping

Incoming data is typed against an economic model — positions, securities, transactions, organizations, investors, markets as interconnected entities — rather than mapped column-by-column against a database schema. Decades of investment in expressive typing and automated tooling for ingesting new sources with new attributes. Less brittle when sources change. Faster to extend.

02

General enginescustomer-owned alignment

The analytical engines accept extensive inputs so they can reason across any problem in a general way. The customer's preferences, constraints, and trade-offs are applied through View on the way in and composed through Narrative on the way out. The institution owns its economic logic; the engines stay general.

03

Whole without seams

Each capability is independently purchasable. The information layer is what holds any combination together — a customer firm can adopt one capability, two, or the whole, and the result has no more seams than a monolith.

04

Reasoning preserved end-to-endbi-temporally

Every decision carries its inference chain. The chain is preserved bi-temporally — what was known when, what was decided when — so any decision can be inspected, audited, or replayed against either its original or current state.

05

Continuousin the way markets actually move

The runtime is batch and event-driven — daily batch, custom-frequency batch, event-driven live — chosen per source by what the source actually produces. Streaming is in development. The pattern matches the cadence of the data, not an arbitrary platform schedule.

06

Outputs in the customer's vocabulary

Decisions arrive in the customer's own data model, not the platform's. Automatic harmonization across systems, security masters, entities, and semantics happens on the platform side, so nothing translates on the receiving side.

07

Open by constructionreach in, don't enclose

ALPHA reaches into the systems the institution already runs instead of asking it to move into ours. Embed the engines anywhere, ingest from any source — 400+ direct connections, 1,500+ via partners — and harmonize every output to the customer's own data model. Interoperability is the default, not a professional-services engagement.

II

Roles

Four boundaries
01

Configurationcustomer owns

Self-service tools. Managed services and channel partners available. High-touch support at scale — direct access to the analysts, economists, mathematicians, and physicists who built the analytics. AI-assisted configuration tooling in development. ALPHA is software and tech, not a services business.

02

Data modelALPHA infers, adapts as needed

The customer doesn't have to declare or document its vocabulary up front. ALPHA infers the customer's data model from its sources and adapts harmonization as the customer's model evolves.

03

DeploymentALPHA cloud · customer cloud · on-prem

ALPHA cloud is the default. Customer cloud is supported. On-prem is available at premium. ALPHA handles uptime and patching on what ALPHA manages; the customer handles what the customer chooses to manage.

04

Auditcustomer pulls

The customer pulls inference chains, lineage, and audit artifacts directly from the system. ALPHA does not gate access to the customer's own decision history.

III

Extensibility

Three modes
01

Bespoke

The customer adds its own data sources, models, preferences, and narrative logic as first-class participants in the architecture.

02

Configured

The customer adjusts ALPHA's existing models and surfaces through extensive parameters and rules.

03

Managed

ALPHA configures and operates on the customer's behalf. Most institutions run all three at once — bespoke where they have a view, configured where they don't, managed where they want leverage.

IV

Operations

Four dimensions
01

Observabilityfull

The customer has access to every data point the platform processes, with lineage attached. Real-time dashboards on the customer side; ongoing monitoring on ALPHA's side. Direct, escalation-tier communication with ALPHA engineering, support, and remediation depending on product configuration. Drill-down from any output to source data, end-to-end.

02

Auditabilitybi-temporal

Every decision is reconstructable. Old versions of models, views, and underlying data are retained — bi-temporally — so the state that produced any past decision can be rebuilt exactly. Past versions persist; decisions remain attributable to the version that produced them.

03

Replay

The bi-temporal architecture is what makes replay native rather than an add-on. Past events can be re-run against past state, current state, or a hypothetical state, with the inference chain preserved at every step.

04

Failure modescontained, surfaced, remediable

Failures are constrained to their specific source. The architecture isolates them rather than propagating them. Status, issue tracking, and remediation are surfaced to the customer in real time; source-level drill-down identifies error sources and lineage without intervention.

Plate IV Surfaces

Three ways the platform reaches the firm.

Through our own applications, through partners, or embedded directly in the firm's infrastructure. The surface is a matter of where the firm already works; the underlying decision is the same.

Native applicationsSurface · Native
ALPHA's own applications, machine-to-machine and human-facing. APIs, file delivery, relational interfaces, direct database integration, and bespoke machine-to-machine connections. Server and cloud-based applications for the people who use them.
In production
Partner platformsSurface · Partner
Software partners with human-facing applications built on ALPHA engines, and distribution partners who embed ALPHA models, analytics, and utilities into their own platforms and marketplaces. Numerous partners including SS&C, BNY, and FactSet.
In production
Embedded — your platformSurface · Direct integration
For institutions running internal portfolio, risk, or research systems, the reasoning architecture can be embedded directly. ALPHA's role is the engine; the surface is your existing infrastructure.
In production
Plate V Specifications

What the system carries.

Selected facts about the platform as it runs today.

What this is
Architecture
Economic operating layer
Not workflow orchestration
Discipline
Decision science
Not operational AI
Reasoning
Probabilistic · causal · adaptive
Inference, not orchestration
Ownership of view
Institution's
Encoded in ALPHA | View
Information
Administered information · AUM serviced
$4T+
PAIX entities, in production; typed, attributed, lineage-preserved
Asset class coverage
Public · private · derivatives
Equity · fixed income · alts
Geographic coverage
Global
Developed and emerging markets
Decision Science
Risk model
Everything Everywhere
Northfield, in production
Methodology depth
40+ years
Continuously developed since 1985
Optimization
Open Optimizer
Constraint-aware, view-aware
Alignment & runtime
Preference modeling
Patented
ALPHA | View · Market76 lineage
Decision cadence
Continuous
Per-event, not per-batch
Auditability
End-to-end
Every decision carries its trace
Surfaces & integration
Native applications
Live
Server and cloud · machine-to-machine and human-facing
Partner platforms
Live
SS&C · BNY · FactSet · among others
Embedded · bespoke integrations
Live
Direct integration to the firm's data model
Delivery formats
API · relational · file · custom
Framed to the receiving system
Institutional deployments
200+
Investors of size
Financial institution deployments
20
Global
Research foundation
Published papers
2000+
Northfield research corpus
R&D facility
ALPHA | Northfield labs
Harvard Square, Cambridge MA
Ownership
Employee-owned, employee-controlled
Structural, not aspirational
Compliance & security
Certifications
SOC 2 Type II · ISO 27001
Independently audited
Disaster recovery & BCP
Full programs in place
Across cloud, private cloud, and on-prem
SLA framework
Calibrated to deployment model
Meets institutional procurement requirements
Talk to us