Applied AI · Payments · Developer APIs & MCP · Partner middleware

I build the core platforms that many products run on.

Applied AI, payments, developer APIs and MCP, partner middleware, ordering, loyalty and CRM, all running on one shared core. At TabSquare AI I owned that core for 8,000 merchants. At Delivery Hero I led 10 direct reports, co-owned the payments P&L across 10+ markets and took a PayFac from $0 to $50M in monthly GMV in six months. Now, through Prodaptix, I ship AI into production for B2B SaaS teams: a multi-agent RAG sales assistant that doubled lead-to-close conversion, and an AI digital worker for refunds and chargebacks. Both keep a person in the loop.

Open to Head of Product and Principal PM roles · Based in Dubai · Relocation or remote across UK, EU & North America

Talk about a role See how it fits together

Ex-Delivery Hero · TabSquare.AI, acquired by Delivery Hero

Monthly GMV within 6 months of launch, from $0: $50M
Merchants on one multi-tenant core platform: 8,000
Partners connected through middleware and open APIs: 50+
Lead-to-close conversion with a RAG sales assistant: 2x

The platform

One core platform. Many products on top.

At TabSquare AI, a customer could place an order from a kiosk, a tablet, their phone, the web, or inside Google Pay and the OCBC Bank app. The order, the loyalty points it earned and the payment that settled it all ran through the same core platform. When the core slowed down, every product slowed down with it.

Applied AI

Recommender systems, ML fraud scoring and multi-agent RAG in production since 2017, shipped with evals, cost controls and guardrails.

2x lead-to-close · 30–40% lower cost per conversation · +8% AOV

Read the deep dive

Payments

PayFac launch, 25+ payment methods, routing across 20+ rails, and reconciliation treated as a product.

$50M monthly GMV within 6 months

Read the deep dive

Developer APIs & portal

OpenAPI contracts, versioning, SDKs and a self-serve developer portal.

3x API adoption · onboarding weeks → days

Read the deep dive

MCP

payments-ops-mcp, an open-source MCP server for reconciliation and chargebacks that recommends but can't move money. MCP permissions in production on the sales assistant.

Open source · 50 tests

Read the article

Partner middleware

Catalog, orders, loyalty and payments synced with partners including Google Pay Singapore and the OCBC Bank app.

50+ partners

Read the deep dive

Ordering

One order flow behind kiosk, tablet, mobile, web and partner apps.

8,000 merchants

Loyalty & CRM

Loyalty synced across every channel and partner through the same core, not rebuilt per channel, plus end-to-end CRM integrations.

Core platform

Shared by every product and channel above

Multi-tenant B2B SaaS
Distributed services
Real-time transactions
Event streams
PostgreSQL read replicas
Redis caching
Observability

Ordering, loyalty, middleware and recommendations: TabSquare AI. Payments, routing, fraud scoring and the API platform: Delivery Hero, after it acquired TabSquare. RAG assistant and the refunds and chargebacks digital worker: Prodaptix client work. payments-ops-mcp: my open-source project.

Products change every quarter. The core underneath decides how fast all of them can move. That’s the layer I lead.

Applied AI in production

Shipping an AI demo is easy. Running it in production is the job.

AI in production since 2017: recommender systems, fraud scoring, RAG and agents. These are the problems I own once it’s live.

01
Evals
A 300-example golden dataset gates every release. It checks accuracy, catches drift when business logic changes, and tests answers against policy for hallucinations.
300-example golden set
02
Cost of AI
Each task goes to the right model: Claude 3.5 Haiku for summaries and after-call work, Claude 3.5 Sonnet for live calls and deal numbers. With bounded context, cost per conversation fell.
30–40% lower cost per conversation
03
Guardrails
Agents can read data but never write it. Refunds and chargebacks need human approval, and the sales assistant's output goes through supervisor review.
A person in the loop on money and customers
04
Grounding
Retrieval only covers the rep's own accounts, and figures the AI extracts are checked against the source before anyone relies on them.
Open-source payments-ops-mcp
05
Memory and context
Session, episodic, semantic and procedural memory, so follow-ups don't start from zero and context stays bounded.
2x lead-to-close conversion
06
Data before models
I design the event tracking before any model is trained, because a model is only as good as the data it learns from.
In production since 2017 · +8% AOV

See the Applied AI deep dive

Selected Work

Selected work

The problem, what I owned, and what changed.

Agent Systems · Memory & Inference Economics

Production Agent Memory & Inference Economics

Prodaptix · Embedded with client · 2024–present

Problem

Stateless follow-ups, scattered account context and repeated retrieval interrupted live sales calls. Unbounded context increased token work and made answers harder to trust.

What I owned

Product ownership of the sales assistant: session history, execution state, episodic, semantic and procedural memory; account-scoped retrieval; MCP permissions; model routing between Haiku and Sonnet; a 300-example golden set; supervisor review.

Company outcome

2x lead-to-close conversion with a RAG sales assistant

Read the full case

Core Platforms · Transaction Reliability

Distributed Transaction Engine & Orchestration

Delivery Hero · Regional Payments · 2021–2024

Problem

Fragmented providers and synchronous dependencies made transaction reliability, safe retries and ledger reconciliation shared product problems across a high-volume marketplace.

What I owned

PayFac product strategy and launch sequencing with engineering, risk and finance: multi-tenant orchestration, idempotent transaction flows, ledger integrity and reconciliation-as-a-service.

Company outcome

$50M

Monthly GMV within 6 months of launch, from $0

Read the full case

Developer Platforms · API Ecosystems

Unifying Developer APIs & Protocol Middleware

Delivery Hero · foodpanda · 2021–2024

Problem

After a series of acquisitions, 10+ markets ran inconsistent APIs and duplicated internal services. Every partner integration was bespoke.

What I owned

API strategy and governance with engineering: strict OpenAPI contracts, versioned interfaces, canonical middleware and shared event streams supporting 50+ partners and client applications.

Company outcome

3x API adoption · onboarding weeks → days

Read the full case

What colleagues say

From people who worked with me on payments, product and platform work.

Payments migration

I have worked with Janahan quite extensively and on one notable project, he was instrumental in ensuring that the transition from one payments provider to another, was timely and smooth. That was an extremely complex task involving a new product, multiple stakeholders, and sensitive and aggressive deadlines. In the first 2 months of the launch itself, the product reached ~$10M in volumes!

Kanika Mittal

VP Fintech & Partner, Travel Payments

Product intuition

What stood out consistently was his sharp product intuition and his ability to craft solutions that were not just functional, but genuinely aligned to market needs. He has a rare ability to listen deeply, to customers, and to the market and translate that into extraordinary product outcomes. He was a true partner in every sense, collaborative, grounded in data, and always thinking a few steps ahead.

Sameer Kaushik

VP Sales, TabSquare AI

Engineering + product

What makes Janahan stand out is his unique combination of technical depth and product vision. With his engineering background, he not only understands what it takes to build stable and scalable products but also brings great ideas that are practical, intuitive, and user-friendly.

Kar How Wong

Regional VP Operations