Skip to content
GroundSetTM
Walkthrough
Platform01Proof02Trust03Company04
Appearance
Sovereign decision infrastructure · Regulated operations

GroundSet is the system of record for enterprise judgement. It reads real-world evidence, applies your rules deterministically, and sets every decision in an immutable audit ledger — inside your environment, on models you own.

Request a walkthroughRead the doctrine
Sites live · India
1,500+
BSP-licensed digital banks
2
Zero retention
In-VPC

Ground + Set

One word, assembled from two.

Ground is what a decision stands on: the evidence in front of you and the rules you wrote. Set is what happens to it next: fixed in the record, where it can be replayed.

01The name0203040506Skip the film
  1. 01The name

    Ground + Set

    One word, assembled from two.

    Ground is what a decision stands on: the evidence in front of you and the rules you wrote. Set is what happens to it next: fixed in the record, where it can be replayed.

  2. 02The parts

    Three parts

    Each one carries part of what GroundSet does.

    Take the mark apart and there are three pieces: a ground, a point and an arc. None of them is decoration.

  3. 03The ground

    Ground

    Evidence and rules.

    Every judgement starts grounded. Real-world evidence is read as it arrives, and your rules are applied deterministically: the same evidence and the same rule version give the same outcome.

  4. 04The point

    Set

    One decision, fixed.

    The point is a single decision. It does not float. It rests on the ground that produced it, and it is set in the decision ledger with its evidence, its rule version and any human override.

  5. 05The perimeter

    Sovereign

    Yours, and drawn open.

    The arc is your environment. GroundSet runs inside it, on models you own. It is never drawn closed: a system of record for judgement cannot be a black box, so every decision can be opened and replayed.

  6. 06The mark

    Three parts. One judgement.

    Grounded. Set. Sovereign.

    Seen from the front, the ground, the point and the perimeter read as a single G. Now you know why it is called GroundSet.

Decision ledger · ReplayEngine active
Example decision record: EVIDENCE IN: handwritten_ledger_0447.jpg · id_scan.png. UNDERSTOOD: monthly_revenue ₱412,300 · 14 debtors · 2 anomalies. RULE APPLIED: MSME-CREDIT-114.v12 · dscr ≥ 1.25 · PASS. DECISION: APPROVE · limit ₱180,000 · 47ms. SET TO LEDGER: 0x8f3a…c41d · replayable · immutable.
  • 01Banking · BSP-regulated
    In production at the Philippines' largest licensed digital bank
  • 02Logistics · Global
    In deployment with the world's largest logistics company — expanding to 2,150 sites
  • 03Physical security · India
    Deployed across three of India's ten largest security networks — 1,500+ sites live
  • 04Insurance · Pan-Asian
    Signed with a leading pan-Asian insurer — claims decision intelligence
  • 05Deployment
    Always inside the customer's VPC. Zero data leaves the perimeter.
  • BSP-regulated
  • MAS-supervised
  • IMDA-aligned
  • India DPDP
  • SR 26-2
  • EU AI Act
  • In-VPC
  • Air-gapped
  • Replayable
01The problem

Every enterprise records everything — except the thing that matters most.

ERP
Transactions
CRM
Relationships
COREBANK
Accounts
HRMS
Workforce
GROUNDSET
Judgement

Human judgement — who approved this loan, on what evidence, under which rule, and why — was never recorded anywhere. It lived in people's heads and died in email threads.

It doesn't scale, either. Every new decision workflow means more headcount, more training, more inconsistency. And the rules engines built to help cannot read the evidence real decisions actually run on: handwritten ledgers, crumpled receipts, scanned IDs, voice logs, camera feeds.

Judgement is the last unclaimed system of record. GroundSet claims it.

02Why now

AI made the gap catastrophic. Regulators made it billable.

4×
Rise in enterprise shadow-AI detections in one year — Verizon DBIR 2026
~⅓
Of organisations report mature governance for agentic AI — McKinsey 2026

Agents now act on core systems. The decisions are happening; the record is not. And the people who present the bill have arrived:

  1. Singapore · Jan 2026

    IMDA Agentic AI Framework

    Every agent must carry verifiable identity and an audit trail of who acted under whose authorisation. The first regulator to mandate the category.

  2. United States · Apr 2026

    SR 26-2

    The Fed, OCC and FDIC replaced fifteen years of model-risk guidance and placed agentic AI outside its scope. Boards must govern it themselves — with an auditable record of every judgement.

  3. European Union · Dec 2027

    EU AI Act, High-Risk

    High-risk obligations for credit and claims decisioning land December 2027. Penalties and transparency duties are already in force.

"The model said so" is not an answer any auditor accepts. GroundSet exists so no institution ever has to give it.

03The platform

Understand. Decide. Route. Set.

One engine, four verbs. Domain-trained models read messy reality; your encoded rules decide deterministically; every outcome routes with justification and is set in the ledger.

Understand

Read the evidence

Domain-trained SLMs and VLMs turn handwritten ledgers, IDs, invoices, voice and video into structured fact. Zero-shot — no template training.

evidence_in: ledger_scan.jpgextracted: monthly_rev ₱412,300confidence: 0.987
01020304
Understand01

Read the evidence

Domain-trained SLMs and VLMs turn handwritten ledgers, IDs, invoices, voice and video into structured fact. Zero-shot — no template training.

evidence_in:   ledger_scan.jpg
extracted:     monthly_rev ₱412,300
confidence:    0.987
Decide02

Apply your rules

Your policies, thresholds and SOPs execute deterministically. Same input, same output, every time. No temperature. No hallucination risk.

rule:          MSME-CREDIT-114.v12
threshold:     dscr ≥ 1.25
result:        PASS
Route03

Act with justification

Auto-approve the clear majority, reject, or escalate to humans with full reasoning attached. People handle true exceptions only.

decision:      APPROVE
route:         auto
justification: attached
Set04

Write the record

Evidence, rule version, outcome, override — written to an immutable, replayable audit ledger. The system of record accretes with every decision.

ledger:        SET
hash:          0x8f3a…c41d
replayable:    true
Explore the platform
04Sovereign by design

Your models. Your perimeter. Your record.

GroundSet runs entirely inside your environment, on open-weight models post-trained in-house and owned by you in deployment. Nothing a vendor — or a foreign government — can switch off.

01Deployment

In-VPC or fully air-gapped. No external calls at inference. Reviewed and passed by Tier-1 bank InfoSec.

02Data

Ephemeral processing. Documents handled in-memory and purged. Zero retention, by architecture rather than policy.

03Record

Every decision replayable end-to-end: the evidence, the rule version, the outcome, the human override. Designed to support BSP, MAS/IMDA, DPDP and EU AI Act requirements.

05What GroundSet is not
  • 01OCR extracts.It doesn't decide.
  • 02Copilots suggest.They aren't accountable.
  • 03Wrappers rent judgement by the token.You never own it.
  • 04Data platforms record context.Not conclusions.
  • 05GroundSet decides —and remembers why.
06Next step

See one real workflow become infrastructure.

We map one decision workflow from your operation and show it running end-to-end — understood, decided, routed, and set in the ledger. Thirty minutes, your data patterns, no slideware.