# Subash Natarajan

> Forward deployed AI architect, Field CTO at SIEL.AI, based in Singapore and London. Takes enterprise AI and application modernisation from pilot to production, working inside the client's teams: strategy with the business, architecture with the delivery teams, code in the repository. In technology since 2009.

Structured version of this page: [profile.json](https://subash.io/profile.json). Contact: suboss87@gmail.com. Also known as Subash N. Last updated: 2026-10-22.

## Quick answers

- Who is Subash Natarajan? A forward deployed AI architect and Field CTO at SIEL.AI, based in Singapore and London, who takes enterprise AI and application modernisation from pilot to production inside client teams across APAC and EMEA.
- What is he known for? Re-engineering a business process before adding AI agents, then proving one workflow in production with evaluations and cost per transaction tracked. Co-author of an InfoQ article on agentic AI architecture for enterprises and of the Multi-Cloud Handbook for Developers (Packt).
- What has he delivered? Measured results such as accounts payable cut from 18 to 3 minutes per invoice, and 34 RPA bots rebuilt on a local LLM layer behind a firewall. Earlier: a Fujitsu Cloud CoE grown to 28 architects across six ASEAN markets, and an MLOps practice at NTT DATA that cut model releases from six weeks to one.
- Where has he worked? SIEL.AI, Fujitsu, VMware, NTT DATA, Logicalis, Ericsson, and C-DOT, since 2009.
- How do you reach him? suboss87@gmail.com or LinkedIn (linkedin.com/in/subashn). Companies that need to adopt AI can apply for his AI Readiness Review at subash.io/assessment.

## Career, one layer at a time

- Now (2024): Enterprise AI in production. Field CTO, SIEL.AI. First idea to production in 8-12 weeks; 3 of 6 pilots live.
- 2023: AI reference architecture. Regional Practice Lead, Fujitsu. Cloud CoE grown to 28 architects across six ASEAN markets.
- 2021: Cloud-native modernisation. Sr. Architect, VMware. Two stalled banking programmes into production within six months.
- 2019: MLOps and team leadership. Associate Director, NTT DATA. Led a 21-person global team; model releases cut from six weeks to one.
- 2010s: Cloud platforms. Cloud and modernisation, NTT DATA and Logicalis. 32 enterprise clients moved onto an NTT DATA cloud platform.
- 2009: Engineering and automation. Engineer, C-DOT and Ericsson. Automated provisioning: new environments in under a day, down from three weeks.

## How he works

Re-engineer the process, then add the agents. Every step of a process goes into one of four buckets (retire, rule, agent, human approval); an agent gets only the steps that need judgement. Six steps: agree the outcome and baseline; map how the work really runs; decide step by step and system by system; prove one workflow in production; set the foundation, then scale; hand over and keep measuring. Rules: no build without a baseline, no go-live without evaluations, cost per transaction tracked from the first slice, every agent has a named human owner, data stays where the client's rules say.

## Selected engagements (clients not named)

- Manufacturing, accounts payable: 18 to 3 minutes per invoice, $7.58 to $2.34 per invoice, field accuracy 65% to 98%, at 6,800 invoices a month.
- E-commerce support agent pilot: 97% fewer model calls; 6 agents per conversation to 1.
- Regulated environment: 34 RPA bots rebuilt on a local LLM layer; maintenance 55 to 14 hours a month; cases needing a person 40% to 18%.
- Knowledge assistant: time to find a policy answer 25 to 3 minutes.
- In progress: on-premises agentic AI for a Gulf bank under SAMA rules.

## Good fit

Shaping an AI or modernisation programme before it is signed, then delivering it; taking an AI pilot to production in a regulated environment; modernising legacy applications with agents; re-engineering a process before automating it; advising leadership on AI, cloud, and modernisation. Not a fit: consumer and mobile products, blockchain and Web3, fundamental ML research.

## Recognition and writing

- [Agentic AI Architecture Framework for Enterprises](https://www.infoq.com/articles/agentic-ai-architecture-framework/): InfoQ, July 2025, co-author ([InfoQ Best Article 2025](https://www.linkedin.com/posts/infoq_softwarearchitecture-infoq-stayahead-activity-7416800144232378368-lzVI)).
- [Multi-Cloud Handbook for Developers](https://www.packtpub.com/en-us/product/multi-cloud-handbook-for-developers-9781804618707): Packt, 2024, co-author with Jeveen Jacob.
- [The New Stack author page](https://thenewstack.io/author/subash-natarajan/): contributing author.
- [fdeops.io](https://fdeops.io/): practical notes on forward deployed engineering.
- Certified: TOGAF, Google Cloud, Kubernetes. AI hackathon winner. Openclaw contributor ([GitHub](https://github.com/suboss87)).

## Pages

- [Home](https://subash.io/): journey, method, engagements, writing, recommendations, contact.
- [AI Readiness Review](https://subash.io/assessment): complimentary, three companies a month, aimed at teams outside software that need to change but lack the budget or the right expert. Application, a 15-minute preparation survey, a 45-minute working session, then a one-page written read within five working days: maturity stage, the one workflow to start with, build/buy/defer, and the next 90 days.
- [Practitioner Toolkit](https://subash.io/toolkit): playbooks on AI strategy, agentic design patterns, LLMOps, and multi-cloud AI.
- [Agentic AI Patterns](https://subash.io/toolkit/agentic-ai-patterns): orchestration, memory, tool use, planning, and reflection for production agents.

## Optional

- [LinkedIn](https://www.linkedin.com/in/subashn/) and [recommendations](https://www.linkedin.com/in/subashn/details/recommendations/)
- [GitHub](https://github.com/suboss87)