Twenty years inside core banking — now building the AI systems regulated institutions can actually trust.
Co-founder of gaigentic.ai. Former T24 implementation lead at Temenos, lead architect at Cloud Lending Solutions, co-founder of Finexus Inc. I pair deep functional ownership of core banking, payments and treasury platforms with hands-on delivery of production-grade AI software.
Independently designed, built and shipped — each one live in production today.
An operating system for ocean-freight forwarders and NVOCCs — ten connected modules from quote to cash to a posted double-entry ledger, all reading from live operational data. A cognitive pricing engine recommends the selling rate with win probability and a protected cost floor; customer-360 intelligence and missing-volume alerts surface declining accounts against their own baseline; an all-in CPQ engine prices ocean FCL/LCL and air with exact, never-floating-point margin; invoicing, receivables and a native GL complete the loop — with a Claude co-pilot that drafts the tedious work, human-in-the-loop on every write.
Sign-up-free, zero-tracking messaging — direct messages, group rooms, a public live feed and secret aliases. Ships a developer SDK, REST API and embeddable widgets, and doubles as a lightweight communication backbone for autonomous AI agents.
A product-marketing marketplace connecting D2C brands with nano-influencers and nearby creators. Brands post “product for content” offers; the platform automates order fulfilment through Shopify and verifies delivered Reels and posts via the Meta APIs using a unique per-campaign code.
A crowd-shopping marketplace pairing buyers in India with verified international travellers carrying spare luggage capacity — retail-price overseas goods, no reseller markup. KYC verification, AML price checks against Amazon, a flat connection-fee model and structured dispute resolution.
An archery scoring and analytics app: log every arrow, then receive analytics that detect form issues, predict the next score and tell a coach exactly what to work on. An installable progressive web app.
An AI intelligence layer for Community Development Financial Institutions. Works alongside an existing loan-origination system — imports a loan tape, interprets Aeris rating data, drafts compliance and board reports, flags at-risk loans early, and answers portfolio questions in plain English.
A privacy-absolute alternative to cloud AI notetakers: no bot joins the call, no audio leaves the Mac. Captures both sides of any meeting — Zoom, Teams, Meet, anything — via a CoreAudio system tap, transcribes live on the Neural Engine, labels who spoke from per-channel signal energy, then a local LLM turns rough notes into documented minutes. Extracts action items with owners and deadlines that carry over into the next related meeting, detects when a call starts and offers to record, and answers questions across every past transcript with timestamped citations.
A multi-tenant sales CRM built on the edge — companies, contacts, activities and a drag-driven deal pipeline with full stage-history tracking, behind session auth with hashed passwords and HttpOnly cookies. Ships a prospect-research pipeline that turns verified prospect lists into a seeded book of business across US, NL and ES markets. Runs entirely on Cloudflare Workers with a D1 SQLite datastore and a React SPA served straight from the worker.
“Your agent signs. Countersign proves it.” An open-source CLI and GitHub Action that makes an AI coding agent’s completion claims falsifiable. Every claim carries the command that would disprove it, and Countersign runs those commands itself; an eleven-rule marker scan, tuned to zero false positives on a 500-file production tree, catches stubs, fabricated data and empty bodies the agent forgot to mention. Turns the agent’s own “all tests pass” message into executable claims, fails the gate when a claim is quietly softened against the base branch, and appends every run to a hash-chained register with a plain-words evidence pack. No model judgement in any verdict. Zero dependencies, so it runs in any CI runner or air-gapped laptop. Apache-2.0, on PyPI as countersign-cli.
Single-tenant, on-premise platforms for regulated institutions, built under gaigentic.ai. Each ships as a one-command Docker stack — the customer owns every container, database and model.
“Claude Code for financial institutions.” Installs onto a bank’s GPU box with one command, samples its schemas and transactions without straining production systems, and maps them to a FIBO-anchored regulatory knowledge base — Basel, ECB, EBA, ESMA, BoE/PRA/FCA and the GCC regulators. It continuously surfaces anomalies, regulatory-threshold breaches and root-cause hypotheses to each role-holder before they ask. Air-gap-safe by default; every claim cited, every decision replayable.
An “all-seeing guardian” for external digital risk: attack-surface scanning, dark-web and underground crawling, brand and typosquat protection, DMARC / SPF / DKIM scoring, phishing and logo-abuse detection, executive-impersonation matching, payment-card leakage detection, third-party vendor risk and automated takedown.
A unified compliance-and-governance platform built as a monorepo of agentic services — transaction monitoring, sanctions screening, case management, risk and control governance, semantic knowledge and audit-explainability — presented behind a single operations console.
An EU sustainability-compliance platform with a single Impact / Risk / Opportunity registry spanning the CSRD, PPWR, CBAM and CSDDD regimes — a unified cross-regulation model that incumbents such as Workiva, Greenomy and osapiens do not offer in one place.
Visa scheme-compliance and acquirer-invoice analysis. Turns the Visa Core Rules and periodic update bulletins into executive briefs and trackable action workflows; reconciles every line of a monthly invoice against the acquirer’s rate card with SQL and LLM reasoning, producing an acquirer-ready inquiry PDF.
An agentic loan-collections system that runs in “shadow mode” — read-only portfolio ingestion alongside a separate operations layer for assignments, communications and audit. Communications are policy-gated and safe-by-default, with role-based access and inbound SMS / WhatsApp / email capture for opt-outs, disputes and complaints.
An agentic origination system for SME/SMB lending. Five autonomous agents — Detective (KYB/KYC, SBA eligibility), Analyst (financial spreading, DSCR with stress testing, dual-ledger variance), Judge (policy engine, deal structuring, ECOA-compliant adverse action), Supervisor (hallucination and consistency checking) and Orchestrator — coordinated through Temporal.io workflows.
Independent verification of what a bank’s finance-AI actually decided. Reads an export of an institution’s monitoring decisions, re-derives from the raw records what its own written policy says should have happened, and reports every place the two disagree — recompute, conformance, timeliness and coverage checks, all deterministic. A domain-blind kernel with six data-only packs: AML transaction monitoring, reconciliations close, regulatory filings, sanctions screening QA, credit decision conformance and ESG reporting figures. Every run writes to an append-only, hash-chained register with SHA-256 fingerprints of every input, reproduces byte for byte, and produces an evidence pack a compliance officer can hand to a supervisor. The engine connects to nothing and shares no runtime with any system it checks; the platform ships as a four-container Docker stack whose data plane physically cannot reach the internet.
Every product on this page was architected and delivered by directing AI coding agents (Claude Code) under banking-grade engineering discipline — production systems built to institutional standards, not prototypes.
The throughline: domain knowledge and engineering decisions sit with the same person — two decades inside financial systems, applied directly to what gets built and how.
Two decades across product banking — implementation, architecture, founding and exit.
Open to BFSI advisory, fractional product and architecture leadership, and selective build work for regulated institutions.