15 months14 year old Agentic AI start-up for BFSI, live in 4 banks.
Production-ready agentic AI inside the bank’s own stack — organised into three suites that map to how a bank actually operates:
Operations Management, Customer Experience & Value, and Data Management.
Singapore
United KingdomSingaporeUnited KingdomThree suites, one AI workforce.
Hiring one of our agents is like hiring a person — job description, salary, supervisor, and everything on the record.
Operations Management
Run the regulated back-officeRepetitive, regulated work off people's desks — maker-checker at every gate.
Buyer · COO, Head of OperationsData Management
Make enterprise data usableTurn enterprise data into a governed, queryable asset with zero PII exposure.
Buyer · CDO, CDAO, CIOCustomer Experience & Value
Grow customer valueConvert existing transaction data into measurable, real-time customer growth.
Buyer · CMO, Head of Retail, CardsLEANM AI.
live at SIB · 6 agents · 7 systems
Thousands of LEA and regulatory notices a month, triaged by hand at ~2.5 hrs each.
Reads, verifies and routes every one in ~30 seconds — one mandatory human gate.
Controls & Anomaly Detection.
live at ADIB
Controls testing samples 2–5% of transactions. Breaches surface days after the fact.
Monitors 100% of transactions, and turns a breach into an alert in under a minute.
Test Data Manager.
synthetic · zero-PII · on demand
Test environments run on copies of production — real PAN, Aadhaar and account numbers.
Generates 100% synthetic, referentially correct test data on demand.
CXO Concierge.
v0 live at ADIB · deterministic plan checker
Every business question becomes an analyst ticket, and a week of waiting for a number.
Ask in plain English and get the answer — a deterministic plan checker, not a guess.
Entity Manager.
live at HDFC + ADIB
Product entities sit in a dozen systems, enriched by hand and inconsistent in every one.
13 agents enrich and distribute them from one source — Shariah checks built in.
AI Marketplace.
live at HDFC · personalised experience
Offers are pushed at segments. Relevance is assumed, and never measured.
65M+ genomes ranked into a personalised decision in under 1.5s — $600M+ incremental spend enabled.
Anticipatory AI.
live demo
Banks compete to acquire the customer, then wait to be told what the customer wants next.
The next phase will be won on anticipation, not acquisition.
Creator Recommendation AI.
proposed · Emtek Group, Indonesia
Creator budgets are committed before anyone knows which creator actually sells which product.
Which creator, which video, which product — decided before the money is spent.
The platform, layer by layer.
What the bank buys sits at the top. Everything below it is shared — which is why a new solution is a combination, not a new build.
Built once. Used by all. Traced always.
Built once
12+ connector classes · one auth modelA new solution inherits the whole fabric on day one — nothing is wired twice.
Tuned, not rebuilt
15 base agents → 270 use casesA use case is a configuration of agents that already exist, not a project.
Traced always
every prompt, tool call and decisionExported to the bank’s own SIEM — the evidence lives on their side, not ours.
The same agents, wherever the data must live.
Cloud-agnostic across every option — one audit trail across all four. Sovereignty is a deployment choice, not a different product.
Local Cloud
Managed by Crayon Data- In-region data centres
- AWS, Azure, GCP and OCI
- Data never leaves the country
Private Cloud
Managed by client- Owned and operated by the client
- Any cloud service provider
- No re-platforming required
On-Premise
Within bank infrastructure- The bank’s own servers and network
- No external connectivity required
- Full control over data, models and AI
AI in a Box
Sovereign · air-gapped- Pre-loaded on hardware the bank owns
- Runs entirely on the bank’s network
- In production in weeks
The right model under the right agent.
Open weights by default, large only where the reasoning pays for itself — and the bar is measured before release.
Versioned prompt libraries, diffable per jurisdiction · ~100 rounds of tuning per bank · regression tests gate every release.Public benchmark, for reference
High performance, versatile, dense model.
30B · 32kBase/Instruct · FreeFast, efficient, open foundation model.
8B · 128kInstruct · FreeBalanced performance, open weights.
12B · 128kInstruct · FreeEfficient Mixture of Experts, versatile.
4B active / 42B · 128kInstruct · FreePer-agent selection · vLLM / NIM · no lock-in to one frontier model
Forward-deployed delivery, productized.
Not a consulting engagement. Fixed outcomes, fixed price, and the RoI modelled before a line of production code — at every phase.
Labs
Discovery & prototypingFoundry
Build & validateFactory
Scale & governOn your data. If the RoI does not model, it stops here.
Signed off by risk and compliance, not just by IT.
Against a held-out control group. You keep the scorecard.
Every engagement returns hardened agents to the catalogue — delivery is also R&D.
Track record
Live, measured, in production.
Everything on the previous slides is running somewhere today — these are the receipts.
LEANM
South Indian BankControls
ADIBAI Marketplace
HDFCCXO Concierge
ADIBCatalyst delivery
KBZ · Riyadh BankSee it running on your data.
This deck is the public overview. Everything else — architecture, working demos, the full capability matrix — sits behind the wall.
Opened under NDAArchitecture, demos and the capability matrix, opened to you.
On your own data, with the RoI modelled before we build.
Thank you.
Scan for the live portal — Singapore · India · Dubai
