IncubatingSport
GimbalSport
gimbalsport.com →A sports venture currently being scoped inside the foundry. Detail and metrics will be published as it leaves incubation.
Case studyVenture studio · Bengaluru · LLPIN ACU-2491
Every enterprise has a system of record. We build the system of judgement.
Agents that decide at a measured error rate and leave a record of every call. Engineers inside your business find the decisions; a factory behind them builds the agents in weeks.
- 2–4
- Weeks to a build
- Every
- Agent decision designed to be recorded
- 4 levels
- Autonomy earned per decision, never assumed
- 5
- Platforms deployed
Why AI stalls in the enterprise
Integrators
Integrators sell certainty: fixed scope and sign-off. They can’t contract for software that is right 94% of the time.
AI startups
AI startups sell possibility: a demo that dazzles, then fails audit.
AI Foundry
We sell governed probability: agents measured on your data, bounded where rules demand certainty, promoted only on the numbers.
How we make AI accountable
Measured. Bounded. Earned. Recorded.
Measured.
Every agent is evaluated on your historical decisions before it touches a live one.
Bounded.
Money movement, filings and system-of-record writes stay deterministic. Agents prepare; rules and people execute.
Earned.
Autonomy rises one level at a time, only when accuracy holds for weeks — and drops automatically if it doesn’t.
Recorded.
Every judgement — inputs, confidence, outcome, override — is written to an open decision ledger you own.
The autonomy dial · earned one level at a time
L0
Observe
Runs in shadow. Nothing it says is acted on.
L1
Advise
Recommends, with confidence and evidence.
L2
Approve
Prepares the action. A person approves.
L3
Act
Acts within bounds. Abstains when unsure.
The catalog · the deployed motion
What engineers inside partner accounts have already built.
Each began as a workflow found inside a business. The platform stays ours; the partner’s instance is theirs. Partners are described by sector only.
LegalOS
first pilotEnterprise
A legal operations stack for in-house counsel — contract lifecycle, obligation tracking, and an auditable trail of who agreed to what, when.
Where judgement is governed:Clause extraction is probabilistic; the obligation record is authoritative.
The legal function of a large diversified Indian conglomerate
LedgerVault
first pilotFinancial Services
The compliance and engagement operating system for accountancy practices — tenant-isolated Postgres with row-level security and document trails built in.
Where judgement is governed:Documents are classified by agents; tenant isolation and the audit trail are enforced by the database, not the model.
An international audit and advisory network
Foreign Remittance Withholding
first pilotFinancial Services · built on LedgerVault
A case workbench for payments to non-residents — take in the document pack, resolve the rules in force on the payment date, characterise the supply, compute the withholding, and prepare the forms the team files by hand. It prepares; it does not file.
Where judgement is governed:It characterises the supply and computes the rate; the team files.
A firm that builds and operates GCCs for US clients
Keel
in buildReal Estate
The operating layer for standing up a Global Capability Centre — talent mapping, workspace, finance and HR run as one instrumented workflow rather than six vendors, so an office park can hand over a running centre instead of a floor.
Where judgement is governed:Workflows are recommended across talent, workspace, finance and HR; commitments go through human approval.
In build for commercial real-estate developers whose office parks host capability centres — not yet in an account
Mould
first pilotEnterprise
A logistics operations platform for asset-heavy networks — movement, utilisation and exception handling against the unit economics of each lane.
Where judgement is governed:Exceptions are scored and ranked; nothing moves an asset without approval.
An industrial logistics group
The portfolio · the cast motion
Spun-off companies.
Standalone companies with their own domains and roadmaps — each in a regulated or high-trust market, each governing where its models may decide.
Sports
Software and infrastructure for how sport is run, coached, and seen.
Pilot · agency partnersSport · Broadcast
NoPixel
nopixelsports.com →Canonical reality infrastructure for sport. Capture reality once as a continuously-updated digital twin; let broadcast, officiating, analytics, betting, fantasy, and coaching all render from one authoritative truth layer.
- ₹4–12L
- Per-event capture cost
- 10–20%
- Of legacy OB cost
- <60s
- Failover
Live in AU & INSport
FlikVault
flikvault.com →Sessions, attendance, billing, reports, parent comms, and AI video analysis — one platform, one source of truth. Built for academies in Australia and India, cricket-first by design.
- 6
- Modules in one login
- 2
- Continents, live
- < 1wk
- Onboarding time
Pilot cohort · 2026 seasonSport
SKrutin
skrutin.cricket →Record on the iPad. AI slow-motion. Ball tracking. Biomechanics scoring. Claude-written coaching note. From bowler’s mark to parent’s phone in three taps — and offline-first at the nets.
- 240
- fps slow-motion
- 3
- Taps to player’s phone
- ∞
- Offline-first at the nets
IncubatingSport · Fitness
The Gauntlet
in the foundryAn owned team fitness league — four-athlete squads representing their school, college, or university compete across a season, anchored by a Bengaluru open-air competition arena. The format IP and the venue, built to travel. (Working name.)
- 4
- Athlete relay teams
- 25k
- Sq ft open-air arena
- School→Uni
- Institutional pipeline
Financial Services
Regulated rails, built for the markets that need them most.
In productionFinancial infra
CortexData
cortexdata.ai →Originate, manage, co-lend, and securitise on a single stack. Five composable modules, eight loan products, every RBI return — built ground-up for Indian banks, co-operative banks, and NBFCs.
- 8
- Loan products on day one
- 5
- Composable modules
- EN · ಕನ · हि
- Languages, equally
Q3 2026 production launchFinancial infra
Hundi
hundi.network →A regulated, blockchain-native settlement venue for the Gulf–India corridor. Atomic settlement in seconds, FATF Travel Rule embedded at the protocol layer, RLUSD and ARC as the on-chain settlement assets.
- 3–5s
- Atomic finality
- 100–150
- Bps blended cost
- $280B
- Annual corridor flow
Pilot · paying-readyFinancial infra
LedgerVault
ledgervault.cortexdata.ai →The compliance and engagement operating system for chartered-accountancy firms — Postgres schema, row-level security and tenant isolation baked in. Five firms ready to pay.
- 5
- Firms ready to pay
- v1.2
- PRD shipped
- RLS
- Per-tenant isolation
Marketing Services
Marketing run by an agent that answers to the unit economics.
Pilot · e-commerce brandCommerce
PRISM
prism.aifoundryventures.com →An autonomous CMO agent for e-commerce — reads live demand, allocates spend against unit economics, and is honest when a channel doesn’t pay. Surfaced a 0.56× ROAS on ₹4.06L of Meta spend in its first audit.
- 0.56×
- ROAS surfaced
- 6
- Command views
- Demand radar
Enterprise
Software for how large organisations actually run.
Private betaWorkplace AI
Breez
getbreez.in →A WhatsApp-native AI chief-of-staff — runs your day from the chat you already live in, on a DPDP-compliant stack with per-user encrypted vaults.
- Native
- DPDP
- Compliant
- GCP
- asia-south1
Media
Perception measured completely, by agents, so people only decide.
Pilot · two deploymentsMedia · Perception
Prahari
coming soonA fleet of agents watches every broadcast channel, in every language, so no human has to. Every mention is entity-resolved and classified for stance against the principal, coordinated narratives are caught across channels, and a decision card reaches the team on WhatsApp within minutes — with a verifiable record of what was covered.
- <2 min
- Broadcast to escalation
- 6
- Languages on the first roster
- ≈99%
- Broadcast noise shed before the model
After the build
Read the full thesis →The partner keeps the system. We keep the frameworks — and the factory keeps growing.
01
The frameworks stay with us.
Evals, the autonomy dial, the compliance spine and the decision record sit under every build. The partner owns its instance; the foundations stay ours and improve with every engagement.
02
Every live use case feeds the factory.
Each deployment runs on real decisions, in real time. What generalises returns to the factory as a measured agent, so the next account starts from evidence rather than from zero.
03
Every judgement is on the record.
We are wiring our agents to write each decision to Warrant, the open-source decision ledger we built and publish under Apache 2.0. Your auditors can read it. So can your regulator.
Read about WarrantWorking with us
Partner with the studio, or put a company through it.
Need an agent? Talk to us — we’ll tell you whether the factory already has one.
Bring a decision your teams make hundreds of times a month. In two weeks, on your own data, we’ll show you whether an agent can take it — and at what accuracy.