From ₹20,000 COHORT 01 · STARTS 1 SEP 2026 Reserve →
·FDE COHORT 01
Enrolling · starts 1 Sep 2026 Reserve a seat →
FDE COHORT 01·Forward Deployed Engineer·BUILD PROOF IN 30 DAYS

AI's best jobisn't building.It's knowingwhat to build.

AI can write code now. The scarce work, a Forward Deployed Engineer's work, is deciding what should exist, earning trust for it, and shipping it into a real workflow. In 30 days you build the proof: an audit, a grounded v1, a controlled action system, evals, rollout notes and a case study.

Duration
30 dayslive · part-time
Live
Tue & Fri7–9pm · 2 a week
Online
₹20,00040 seats
Studio
₹30,00030 seats · weekly desk
Reserve a seat → Read the method ↓

* The workflow above is real. We rebuilt it at 7th Pillar. Watch it get audited. →

Workflow audit● Diagnosing
enquiryinbound draft quoterepetitive approvejudgment send + logrepetitive follow-uprepetitive automate 80% ← keep human FINDING · ROI grunt work 4 of 5 steps reclaimed 0 hrs/wk verdict SHIP IT
The audit is the job. Circle the repetitive work, keep the judgment call human, price the win. Only then do you build.
01The opportunitythe market for the role

The role AI teams are hiring for, and still struggle to find.

Palantir pioneered the Forward Deployed Engineer. Now frontier AI teams and vertical AI startups want the same kind of person: someone who can sit inside a business, map the messy workflow, scope the right system, and ship it with the customer.

The top of the market chases rare engineers with this mix of judgment, code and customer trust. You do not need to be senior to start building the proof. You need one real workflow, a tight scope, working software, evals, and a case study that shows your decisions.

Market · Forward Deployed Engineer
demandfrontier teams + vertical AI startups
supplyrare mix of code + customer judgment
hiring signalproof of deployed judgment

We can't promise you a salary or a title. We can teach the work hiring teams scan for, and send you out with proof you can defend.

Scarce skill
02The two lenses · methodwhat you become
01 The consultant's eye

Read the business.

Before a line of code, you learn to walk into a business and see it the way an operator does.

01Interview both levels, leadership for the goal, the people doing the work for the reality, and map how it actually happens.
02Classify every step: plain software, an agent, or human judgment.
03Find the grunt work and the bottlenecks, where time and money leak.
04Spot where AI will quietly break, and where it's too risky to try.
05Model the ROI honestly, then rank every win by impact against effort.
06The hardest discipline: knowing where AI does not belong, and saying so.
02 The engineer's hands

Ship the system.

Then you prove the diagnosis the only way that counts: you build the AI agent and ship it into production.

01Prototype the AI agent from a tight spec: build, test, review.
02Ground it in real data with RAG, knowledge graphs, so it stops making things up.
03Make it act with tools and MCP, under human-oversight gates.
04Write the evals that prove it works. Watch the cost.
05Deploy it, callable by other software.
06Ship a case study you can defend to an engineer and a CEO at once.
Two lenses, one person. That's rare, and it's what businesses can't hire enough of.
03The weekly rhythmhow it runs

What getting it into the open actually looks like.

Two live sessions a week, build in between, ship every Friday. Both 7–9pm, so you don't quit your job: Tuesday we brief the project, Friday we review and ship what you built. Online, you build on your own time; Studio, you build at our Kochi office next to Rashin. The WhatsApp room stays open between sessions.

Tuesday · live
The brief
We brief the project live, 7–9pm. It is recorded, but the point of live is simple: you ask in real time.
Remote
Between · your hours
Build it
You build the project on a real system: audit, prototype, agent, evals. Online, on your own time; Studio, at our Kochi office next to Rashin.
Remote · or Kochi
Friday · live
Ship review
We review what you built and ship it, 7–9pm. Working code, not slides.
Remote
Between sessions
The WhatsApp room
The cohort and Rashin stay in one thread. Post the blocker, the repo, the screen. You keep moving.
Async
04The WhatsApp roomthe community

You're never stuck alone.

The cohort runs 30 days. The room is why people finish. One WhatsApp group: the whole cohort and Rashin, running from day one.

  • –Blocked between sessions? Post the failing run, the screenshot, or the repo. The room has context.
  • –Drop your business audit in, get it torn apart, kindly, before you build the wrong thing.
  • –Share the win the second it ships. The room keeps you moving.
  • –It outlasts the cohort. The network is half of what you're paying for.
05Work plan · week by weekone real agent

Diagnose first. Then build the cure.

30 days, four weeks, one thing built end to end: an AI agent that does a real job. Week 0 gets you to the starting line if you've never shipped an AI product; if you have, it's a speed run. The rhythm stays the same every week: Tuesday's brief, Friday's ship, WhatsApp in between.

The curriculum stays live. AI moves fast, so each cohort we adapt the projects, tools and tech to keep you on what actually ships today, not what was true last year.

00
Week 0 · Setup

Get AI-product ready.

Before the first live class, you set up the baseline stack and build the smallest possible AI app. This keeps juniors from drowning in tooling, and gives experienced devs a clean runway for the real project.

What it covers
  • API keys, environment variables, local setup, deployment account, and safe secret handling
  • One LLM call, one structured JSON response, one schema validator, one failure path
  • A tiny CRUD app with auth-shaped thinking: users, records, permissions, logs
  • Agentic coding workflow: spec, agent run, review, tests, commit
  • The minimum git and PR hygiene you need before AI starts writing code with you
DeliverableA tiny AI CRUD app, deployed
You build

A small deployed app that calls an LLM, validates structured output, stores a record, and logs the run. It is not impressive. It is the floor.

Tools

Agentic coding · Next.js · API keys · git · deploy preview

01
Week 1 · Read

Audit a real business, pick the agent to build.

You start where a Forward Deployed Engineer starts: with the business. We hand you a realistic business to audit, its goals, its operators, and its messy workflow. You learn enough about LLMs to judge them, take that workflow apart, interview the people who run it, and choose the one agent worth building first, with evidence.

What it covers
  • LLM basics for judgment: tokens, context, cost, latency, and why models hallucinate
  • Run discovery interviews: leadership for the goal, operators for the reality, and the exceptions nobody documents
  • Classify every step: deterministic software, an agent, or human judgment
  • Find the bottlenecks and the grunt work, and mark where AI will quietly break
  • Write success criteria, acceptance criteria, and the first golden dataset v0
  • Model ROI: cost cut, revenue freed, capacity created, and the cost of being wrong
  • Rank wins on impact, effort, risk and trust. Say where AI does not belong
DeliverableA workflow audit + the agent you'll build
You build

A written audit of the realistic business we give you. Every step is classified, the first move is scoped, and the first 20 test cases define what “works” means before you build.

Tools

Claude · discovery interviews · workflow mapping · ROI model · golden dataset

02
Week 2 · Build

Build your agent, grounded in real data.

Turn the diagnosis into a working agent. You build it with an agentic coding workflow, give it a real endpoint, and ground it in the business's own data so it answers from evidence instead of guesses.

What it covers
  • AI-native development: spec-driven building with context, subagents, tests and review
  • Calling the LLM APIs: system prompts, tool use, streaming responses
  • Structured output with schema validation, so the format can't drift
  • Retrieval (RAG): embeddings, pgvector, chunking, hybrid search, citations
  • Grounding on real documents: contracts, PDFs, tickets, not toy data
  • Eval v1: source faithfulness, answer quality, empty-context behavior, and log capture
  • Basic production hygiene: secrets, rate limits, retries, and safe error messages
DeliverableA working, grounded agent that does one real job
You build

A live agent that does one real job, answers from real data with citations, logs every run, and passes the first eval gate.

Tools

Agentic coding · Next.js · Postgres + pgvector · embeddings · schema validation

03
Week 3 · Act

Give your agent tools and autonomy, under control.

An agent that only answers is still a demo. You make yours do work, earning autonomy step by step: deterministic workflow first, tool-using agent next, open-ended autonomy only where the workflow needs it.

What it covers
  • Workflows vs agents: when fixed code paths beat open-ended autonomy
  • Tool calling, multi-step planning, MCP, and clean tool documentation
  • Agent harnesses hands-on: Claude Agent SDK, LangGraph / DeepAgents
  • Browser automation and voice as optional add-ons, only where the workflow needs them
  • Oversight gates: draft, ask, approve, execute, roll back
  • Checkpointing, retries, failure recovery, and state
  • Observability: trace exactly what the agent did, and why
  • Security lab: prompt injection, tool misuse, least privilege, PII, output validation
DeliverableAn agent that takes action, under control
You build

Your agent takes real action on the workflow, with human-approval gates on risky steps, scoped permissions, recovery when something fails, and an audit trail you can inspect.

Tools

Claude Agent SDK · LangGraph / DeepAgents · MCP · Playwright · Vapi optional

04
Week 4 · Prove

Prove it, deploy it, roll it out.

The Forward Deployed Engineer's real work is getting the agent trusted. You prove behavior with evals, put it behind production controls, and write the case study a technical lead and business owner can both read.

What it covers
  • Evals at depth: a golden dataset of normal, edge, ambiguous and high-risk cases
  • A failure taxonomy and pass/fail thresholds you can defend
  • Regression runs on every meaningful change, with human judgment calibrated against automated grading
  • Cost tracking, budgets, latency, rate limits, and guardrails against runaway spend
  • Deploy to production or a controlled pilot; expose it as an MCP server when that is the right integration
  • Monitoring in production, catch drift and regressions before users do
  • Rollout plan: owner, users, training notes, acceptance criteria, support path and rollback
  • The case study: diagnosis → build → evals → ROI, in one document that reads to an engineer and a CEO
DeliverableEvaluated, deployed, documented
You build

The system deployed or piloted, an eval report, a runbook, and the case study. Together they show how you think, not just what you built.

Tools

Evals · cost tracking · monitoring · MCP server · auth checklist · runbook

Optional deep-end: fine-tuning, voice, browser automation and multi-agent setups stay optional. Most Forward Deployed work is prompting, grounding, evals, integrations and trust, so the course keeps those in the center.

06On completionwhat you can do
Run an AI audit of any business's workflow.
Tell them where AI belongs, and where it doesn't.
Scope, sequence and price the work.
Build it, ground it, and give it oversight.
Prove it works with evals, logs, and cost controls.
Ship with security, permissions, rollout and rollback in mind.
Defend it to an engineer and a CEO in the same room.
07Your instructorwho's teaching
Rashin Pothan, founder of 7th Pillar Infotech and REthink AI
RASHIN POTHANSINCE 2014
CEO 7th Pillar Infotech (Kochi) · Founder & CTO REthink AI (Arizona, US)

I've been shipping software since 2014. Production AI for the past three years.

I didn't read about this role. I do it. I started with our own shop: I audited how 7th Pillar itself runs (leads, follow-ups, proposals, contracts, marketing), decided where AI belonged, then built it on LangGraph and DeepAgents, with the boring parts that make it trustworthy: oversight gates, eval runs, cost tracking, an MCP server. That's both lenses: read the business, then ship the system. Then I ran the same audit-and-ship for client businesses, not just our own.

I also founded REthink AI in Arizona to build AI for field service businesses: voice agents that answer calls, qualify leads and book jobs, plus the dispatch and follow-up automation behind them, running live 24/7. My team ships to paying clients on four continents. None of us learned this in a classroom, and I'll teach you the way we actually learned it.

RashinCEO ·
7th Pillar Infotech
12yrs running 7th Pillar
4continents served
2AI companies in production
24/7voice agent live in the US
We ship this, we don't just teach it

Products built and run by the people teaching you.

The same audit-to-agent work you'll learn in the cohort, shipped as real products at 7th Pillar and REthink AI.

FunnelFab funnel builder, marketing landing
FunnelFabfunnelfab.com ↗

Build, validate and ship interactive marketing funnels straight from your AI assistant, over MCP. It asks the right questions, sorts the good leads from the rest, and follows up on its own.

Next.js · MCP · Cloudflare
BookingOps operations dashboard
BookingOpsbookingops.app ↗

A unified operations platform that connects admins, service providers and customers across web and mobile, so bookings, schedules and jobs all run in one place.

NestJS · Next.js · Expo
Catava AI receptionist dashboard
Catavacatava.buildmatic.ai ↗

A multi-channel AI receptionist for phone, chat, email and WhatsApp. It books appointments, takes orders and answers questions for any business, set up with a 4-step wizard.

Next.js · Pipecat · Supabase
08Enrolmenttwo ways in
FDE · Online Cohort 0130 days · 2 live sessions / week
Tier 1
Online
₹20,000
incl 18% GST · 40 seats
  • –All 8 live sessions, Tue & Fri, 7–9pm
  • –Every project, guided end to end
  • –Session recordings to keep
  • –The cohort WhatsApp community, from day one
  • –Your deployed system + a case study that proves the work
Reserve · Online →
Tier 2
Most hands-on
Studio
₹30,000
incl 18% GST · 30 seats
  • →Everything in Online, plus:
  • →A desk at our Kochi office one full day a week (9am–5pm), 4 build days in all, capped at 12 a day
  • →Build alongside Rashin, in the room, before each Friday ship
  • →Closer project review, and first dibs on office slots
Reserve · Studio →

Secure checkout via Dodo Payments · Non-refundable · For developers, 1yr+ JS or Python · AI product experience not required · Questions? WhatsApp +91 88481 03435

09Questions · straight answersbefore you reserve
Q1What's the difference between Online and Studio?
Both tiers get the full cohort: two live sessions a week, every project, recordings, the WhatsApp room, and a deployed system with a case study. Studio adds one thing: a weekly desk at our Kochi office, where you build alongside Rashin before each Friday ship. Online is ₹20,000, Studio is ₹30,000.
Q2What if I'm not in Kochi?
Take the Online tier. The whole cohort runs remotely: both live sessions, your build time, and the WhatsApp room. The Kochi office desk is the one thing only Studio gets, and it's only worth it if you can come in. Not in Kochi? Online is built for exactly that.
Q3How does the Studio office desk work?
Studio students get a weekly desk at our Kochi office: book one full day a week, 9am–5pm, whichever day suits you (4 build days across the cohort, capped at 12 desks a day). You come in, sit with Rashin and the other Studio students, and build before each Friday ship. It's the Studio-only part; Online doesn't include it.
Q4Do I need to know how to code?
Yes: at least a year of JavaScript or Python, and you can build a basic CRUD app on your own. The new skill here is the business side, reading a workflow and deciding where AI belongs, which most developers have never been taught.
Q5What if I've never built an AI product?
That's fine. Week 0 gets you through the basics before the first live class: one LLM call, structured output, schema validation, a tiny CRUD app, deployment, and the AI coding workflow. You should know web development; you do not need prior RAG, agents, MCP, or evals.
Q6Why teach the business audit? I just want to build.
A developer who only builds waits for someone else to choose the problem. A Forward Deployed Engineer can walk into a messy workflow, find the useful AI system, price the win, name the risks, and then build. That judgment is the course.
Q7Do you teach security and safe deployment?
Yes. The course covers prompt injection, tool misuse, least-privilege permissions, secrets, PII, output validation, rate limits, logging, human approval gates, rollback, and cost controls. You won't ship agents with blind trust.
Q8Will this make me a Forward Deployed Engineer?
It gives you FDE proof-of-work, not a guaranteed title. Senior FDE roles usually expect years of engineering and customer-facing experience. This cohort gives junior developers a credible artifact: audit, build, eval report, controlled deployment, rollout notes, and a case study you can defend.
Q9What do I walk away with?
A deployed or piloted AI system built on a real workflow, an eval report, a runbook, a case study, the audit skill you can reuse on any business, and a cohort network that outlasts the 30 days.
Q10Refunds?
Seats are non-refundable. It's a small cohort and reserving holds your place, so please be sure before you book. Questions first? WhatsApp us. We'd rather you ask than guess.
FDE · Online Cohort 01 · 30 days · 2 live sessions / week

Find where AI belongs. Then ship it.

30 days, one real system you deploy or pilot. Live online Tuesday and Friday, the WhatsApp room in between, and a weekly Kochi office desk on the Studio plan. Online ₹20,000, Studio ₹30,000. When the seats are gone, they're gone.

Choose your plan → Questions? WhatsApp us →