Manish Shetty

Manish Shetty

AI Deployment and Automation Lead

I build AI systems and automations by directing AI coding agents (Claude Code, Codex), and I make them safe to run inside a business.

I've also built a 200+ client practice, sold technical products in the Gulf and led a remote team of up to eight, so I'm as comfortable in front of customers as I am in the build.

  • Bengaluru, India
  • Open to remote work and relocation for the right opportunity
  • Available immediately
Start here: how I run AI agents5:04

Five systems, built with AI coding agents

Working systems for sales research, payments, operations, monitoring and sensitive health data. Each was built by directing AI coding agents, with automated tests, a release gate and my sign-off.

AI tools for day-to-day operations3:05 walkthrough

Internal AI tools

Three tools the business runs on every day: an outreach engine that only writes from verified evidence, a report generator that won't render without a named person's sign-off, and a content engine that plans, checks and produces daily posts and short videos.

15 of 15

planted errors caught by the report generator's 13-rule claim checker

3 of 3

unsigned reports refused: nothing renders without a named sign-off

0

repeated topics across 150 content draws from a 506-tag bank

Built with Claude Code, Codex, Python, Node.js, Remotion, Playwright

Also rebuilt the company website: Lighthouse 100 on desktop in all four categories.

AI sales-research pipeline2:41 walkthrough

Mandate Radar

Turns a news event into a ranked sales lead with a researched contact and a draft message, and can't save an unsourced fact.

about USD 0.11

per completed run, under a hard daily spending ceiling

290 of 290

stored facts carry a source and a verbatim quote

under 1 min

median, from news event to contact research

Built with Python, Anthropic API, Serper search, SQLite

Read the case study

Sales, delivery and lifecycle-automation platform3:54 walkthrough

Sales and lifecycle platform

A sale is taken, checked, fulfilled and followed up with no manual step.

about 1,900

automated tests behind one release gate

66

database test suites in the same gate

6

defects found in a pre-launch review, each fixed the same day with a regression test

Built with Stripe, Supabase, Resend, Meta Marketing API, MCP, Vercel

Read the case study

Execution-monitoring SaaS3:53 walkthrough

Forge

Turns a two-minute daily check-in into 8 metrics and early-warning alerts for a human admin, who decides what to do.

about 10 weeks

to version 1 of a 57-table SaaS, against a 14 to 21 week plan

about 2,000

automated tests, in gated phases closed on my sign-off

0

false status flips in testing, against 13 for a simple rule

Built with Next.js, TypeScript, Supabase, Vercel

Read the case study

90-day programme platform for sensitive health-related data3:46 walkthrough

Transformation Accelerator

Data protection sits in the database itself, where an app bug can't bypass it.

71 of 71

tables under access control

2

security defects caught early by the agents' stop-and-flag rule

1,425

database checks across 36 test files

Built with Next.js, TypeScript, Supabase (EU region)

Read the case study

How I run AI agents

I set what each system must do and must never do. The agents draft the specifications and code, and I review, test and sign off every release.

  1. A written rulebook per project

    Read at the start of every session: what the product is for, the rules it must never break, which decisions belong to a person, and each past incident with the rule it produced.

  2. Human approval for anything consequential

    Database and security changes, user-facing wording, moving to the next phase, deploys, ad changes and any spending. Approval is per action.

  3. Specs frozen before code

    Work moves in small phases, each opened from a written spec and closed at a tagged version after testing.

  4. Refuse rather than guess

    Agents are told a correction is worth more than compliance. On one build, those stops caught errors in the spec and two real security defects.

  5. Rules become checks

    A rule that lives only in a document gets forgotten, so where possible it becomes a test, a gate or a database rule.

  6. Check reality

    Look at what production serves, query the database and render the page, whatever the agent reports.

  7. A separate review before launch

    By a different session, before real money or real users touch the build.

From the incident log Added an ask-before-any-paid-call rule after an agent made 89 unapproved calls.

Read the method, with the incidents behind each rule

Experience

  1. 2021 to present

    Founder; AI Systems and Automation Lead

    Phalanx Labs, Remote

    • Built the company's AI systems and automations by directing AI coding agents (Claude Code, Codex): I set what each system must do and must never do, the agents draft the specifications and code, and I review, test and sign off every release.
    • Built a 200+ client practice from zero through long-form content and a 2,000+ subscriber newsletter, selling to and onboarding every client myself and delivering against each client's own goals.
    • Led a remote team of up to eight since 2022.

  2. 2018 to 2021

    Graduate Entrepreneurship Programme

    Birmingham City University, Birmingham, UK

  3. 2017 to 2018

    Engineer, Technical Bids and Vendor Management

    Aston UK International Ltd, Birmingham, UK

    • Wrote technical bids for Fortune 500 clients and managed vendors.

  4. 2015 to 2016

    Technical Sales Engineer

    Aatiq Al Dhaheri & Co. LLC, Abu Dhabi, UAE

    • Sold about AED 1 million of technical products, set up an OEM supply partnership and ran product demonstrations at ADIPEC.

  5. 2015

    Mechanical Engineer

    Contract Resources Petrochemical Services, Gujarat, India

  6. 2014 to 2015

    Computational Design Intern

    Mercedes-Benz Research and Development India, Bengaluru, India

Skills

AI and automation
Claude Code, Codex, Anthropic API, OpenAI API, LLM workflows, AI agents, prompt design, model routing, MCP, workflow automation, Make, Zapier
Integration and platforms
REST APIs, webhooks, Stripe, Supabase, Resend, Meta Marketing API, Vercel
Delivery and quality
Requirements and acceptance criteria, automated testing, release gates, AI evaluation (evals), cost control, human approval steps, privacy by design
Customer and team
Discovery and scoping, solution design, demos and proofs of concept, proposals, implementation, client onboarding and training, stakeholder management, team leadership

Education

  • MSc Enterprise Systems Management

    Birmingham City University. 2.1 with commendation, 2017 to 2019

  • B.E. Mechanical Engineering

    Visvesvaraya Technological University. First class, 2010 to 2014