Who you hire.
Identify candidates most likely to reach the role's defined performance outcome.
Hiring proven in productionPeople decisions
The People Decision Engine
Don't trust our AI. Trust your own outcomes.
Every people decision costs both ways: the risk a process accepts and the talent its filters never surface. Nodes tests both against your own history, then returns the evidence and limits before the next decision. You define the outcome and policy boundaries. A named reviewer makes the call.
Read-only history. No live decisions or workflows changed.
What Nodes is
Your outcomes become evidence for the next decision.
One decision engine, four decision programs. Each is trained and validated separately on your outcomes, inside your environment. No decision-specific score or validation result transfers between programs.
Who you insure.
Replay past underwriting decisions against the carrier's approved risk and loss outcomes.
Separate historical validationWho you fund.
Replay past credit decisions against the institution's approved repayment outcomes.
Separate historical validationWho you admit.
Replay past admissions decisions against the institution's defined student-success outcome.
Separate historical validationThe problem
The decision and the outcome live in different systems.
The source system records who moved forward.
Your hiring system stores who was hired. Your policy system stores who was insured. Your loan system stores who was funded. Your student system stores who was admitted.
Performance, claims, repayment, and student outcomes appear later, usually in different systems.
When those records stay apart, a filter can remain in place without ever being tested against the result it was meant to predict.
Nodes reconnects the decision, evidence, reviewer, and result.
How it works
You define the outcome. Nodes shows what the evidence supports.
An illustrative decision, end to end
Signal
A top producer's CRM activity drops. HRIS engagement dips with it.
Draft
A retention play is proposed: comp review, rebalanced book, ROI attached.
Approval
The director reads the Decision Trace, adjusts the timing, approves.
Writeback
Calendar, pipeline, and comp review update. The reasoning writes back too.
Feedback
The outcome is recorded. The graph learns which plays retain.
Candidates run in shadow and require named-human promotion. Each measured outcome becomes evidence for the next governed evaluation.
Security and deployment
Your production data stays in your cloud.
Nodes puts production inference inside the customer environment. Your team retains the boundary and review path.
VPC-resident and single-tenant
Production data, deployed customer weights, and Decision Traces stay inside the approved environment.
No hosted model call
Open-source foundation models run inside the VPC. No hosted model provider sits in the production data path.
Signed, queryable Decision Trace
Each approved action carries the evidence and human input behind it.
SOC 2 Type I and Type II
The held reports cover the Nodes control environment. Scope and artifacts are available for review.
Integration speed has a mechanism.
Templated connectors map fields against a canonical glossary with confidence bands. Anything uncertain goes to a human for review. A changed field stops the flow instead of being guessed at.
One reference deployment.
At the reference carrier, legal approval took 17 days after six AI hiring vendors had been rejected over 18 months on architecture. Contract to production took 34 days. These are observations from one deployment, not a timeline promise.
First production proof
The first proof came from enterprise talent.
At one Fortune 500 insurance carrier, Nodes connected four years of hiring records to actual post-hire production.
Days from requisition to hire across the whole loop.
A 47-day reduction to the first production milestone.
Q1 net savings, CFO-validated at the reference carrier.
Retrospective filter audit
Historical replay found 2,863 producing agents the insurance-experience filter would have rejected.
Those producers represented $17.7M in annual production at risk under that filter.
How these numbers are classified.
The time and savings figures are observed results at the reference carrier. The 2,863 and $17.7M figures are a retrospective counterfactual on observed production. Results come from one customer and are not projections.
A new company or decision program must establish its own evidence through historical validation before live use.
Human control and fairness
A person is not a score.
Nodes evaluates one person for one decision, one context, and one defined outcome.
Each score stays bound to that decision, context, and outcome. It never follows someone across roles, companies, employment, lending, insurance, or admissions.
A named person reviews the evidence and makes the final call. Nodes cannot make an automated adverse decision, carry a score across contexts, or take consequential action without named human approval.
Fairness needs a record.
Historical data can carry historical bias. A replay tests rules and filters against measured outcomes and group-level results before live use. It can surface a difference that needs investigation. It cannot prove a process is fair.
Your people, legal, and data-governance teams define protected-data handling, comparison groups, thresholds, review paths, and corrective steps.
Each approved action creates a signed Decision Trace with the evidence available, limitations, model and policy versions, human input, approved action, and later outcome. The trace supports review without replacing job-related validation, adverse-impact analysis, or legal judgment.
Ownership and portability
Your decision history outlasts the model.
Models change. Your outcomes, Decision Traces, calibration history, and approved policies remain customer-controlled.
Nodes runs inside your VPC. Your organization controls its production data, deployed model weights, and traces. The executed agreement defines export, transition, and exit rights.
The underlying model can be replaced without discarding the decision history built from your outcomes.
Built from your outcomes. Governed by your people. Run in your cloud.
Decision Replay
One decision. 72 hours. $1M identified. Or $0 in production fees.
Give us one high-stakes people decision, such as who to hire, insure, fund, or admit, and at least two years of linked decisions and measured outcomes. Before the clock starts, we agree on sufficient data volume, the required data quality, and a written $1M evidence bar approved by your finance team.
The 72-hour clock starts only after the agreed deidentified dataset is delivered under a signed mutual NDA and accepted by Nodes. Nodes runs the replay read-only inside the approved environment. No live decision or workflow changes.
Within 72 hours, Nodes shows whether your history clears the written bar. That finding is historical evidence, not realized savings. If the history falls short, you owe no production fee and keep the analysis.
- 01
Decision Replay
Test one historical decision against one defined outcome. Past data only. No live operational changes.
- 02
Shadow Mode
Run Nodes beside the current process. Compare recommendations case by case while the current workflow stays in place.
- 03
Controlled Deployment
Go live for one approved decision program. A named person approves every consequential action.
Hiring is proven in production at a Fortune 500 insurance carrier.Other programs begin with their own historical validation.
Request a Decision Replay