Build Fast with AI
CurriculumFormatFrom the JD
Enrollments open·November batch

Forward Deployed
Crash Course

An intensive crash course in everything FDE. All in two weeks.

What you get
  • Live sessions with SatvikLive teaching, doubt clearing and your capstone review
  • 4 hands-on client buildsMulti-agent, MCP, human in the loop, evals
  • 1 final capstoneA client brief, shipped end to end in Week 2
  • 10+ tools, hands-onClaude Code, LangGraph, CrewAI, MCP and more
  • 1 year of course accessCome back to every session and build
01 / Curriculum

The Complete FDE Playbook

Learn live. Build daily. Ship a capstone.

Module 01–02

Learn it live

Module 01 · Live

Build agents from scratch

Write the agent in plain Python before you touch a framework, so you know what every framework does for you.

  1. 01The FDE workflow

    From a client brief to a technical spec.

  2. 02Tool calling

    Tool schemas, structured outputs, and the agent loop.

  3. 03Agent harness

    Context engineering, state management, stop conditions.

  4. 04RAG and agentic retrieval

    Embeddings, vector search, chunking, reranking.

  5. 05Agent patterns and human in the loop

    Router, orchestrator-worker, ReAct, reflection, approval gates.

  6. 06Agent frameworks

    LangGraph, CrewAI and Agno. How to choose one and build with it.

  • Python
  • LangGraph
  • CrewAI
  • agAgno
Module 02 · Live

Agent-driven development

Ship real software the way FDEs do, with coding agents writing most of the code and you reviewing all of it.

  1. 01Agentic coding with Claude Code, Codex and Cursor

    Spec-driven development, plan mode, diff review.

  2. 02Claude Skills, subagents and agent loops

    Reusable workflows and long-running tasks.

  3. 03MCP (Model Context Protocol)

    MCP servers and integrations with Slack, GitHub and Notion.

  4. 04Claude Managed Agents

    Hosted agent runtime and sandboxed execution.

  5. 05Evals

    LLM-as-judge, test datasets, regression testing.

  6. 06Guardrails

    Prompt injection defence, PII handling, output validation.

  • Claude Code
  • Codex
  • Cursor
  • MCP
  • Slack
  • GitHub
  • Claude
  • pfPromptfoo
Days 3–6

Build it daily

4 client projects
Tool calls
→ google_search("NVIDIA news") ✓→ get_stock_price("NVDA") ✓→ notion.create_page ✓Morning report saved to Notion, as JSON.
Build 01 · Day 3

Build a tool-calling agent

An agent that calls live APIs and your own functions, returns clean JSON, and connects to the client's apps over MCP.

Client brief

“Our analysts check stock prices and news by hand every morning, then paste it all into a report.”

You hand in

A recorded run with every tool call shown.

Retrieved chunks
Can we cancel 20 days before?
Cancellation and refunds · §40.82
Children and infants · §20.71
Booking and payment · §30.64
50% refund · cited §4
Build 02 · Day 4

Build a RAG agent with citations

An agent that answers from the client's documents and cites the exact section for every answer.

Client brief

“Our team answers the same policy questions all day from 40 PDFs, and customers don't trust an answer without a source.”

You hand in

A recorded run that shows the answers and their citations.

Agent graph
New ticket↓Router↓
BillingShippingRefunds
Build 03 · Day 5

Build a multi-agent system

A support system where a router sends each ticket to the right specialist agent, and a checker reviews every reply before it goes out.

Client brief

“Every ticket lands in one queue. Billing, shipping and refunds all wait on the same three people.”

You hand in

A working agent graph and a one-page architecture note.

Eval report
Hallucination checkPass
Context relevancePass
Guardrail · personal dataPass
Prompt injectionFix needed
Build 04 · Day 6

Build evals and guardrails for your agent

A golden test set, guardrails for personal data and prompt injection, and a trace of every run in Langfuse.

Client brief

“Before this goes live, show me it won't make things up or leak customer data, and that we'll know when it breaks.”

You hand in

An eval report, a trace link and the known limits.

Day 7 → Week 2

Final capstone

1 client · built end to end

Pick a client. Ship the system.

Choose a brief on Day 7 and spend Week 2 building it. Every build gets reviewed.

Pick your brief
  • Claims triage agentInsurance
  • CRM copilotSales
  • Contract reviewerLegal
  • Bring your ownYour call
You ship
  • GitHub repo
  • Scoping doc
  • Architecture doc
  • Eval report
  • Handover README
  • 5-minute client readout
Week 2, day by day
  1. Day 7
    Scope itPick the brief and write a one-page scoping doc.
  2. Days 8–10
    Build the core systemAgents, tools and integrations, built with coding agents.
  3. Day 11
    Mock decomposition roundBreak down an unseen client problem, interview style.
  4. Days 12–13
    Harden itEvals, guardrails, tracing, and the handover docs.
  5. Day 14
    Ship and presentDeploy it and record a 5-minute client readout.
Eight skills your capstone proves
  1. Agent designAgents built from scratch, in code
  2. Multi-agent orchestrationRouters, orchestrators and specialists
  3. Tool use with MCPAgents that act inside real systems
  4. Coding-agent workflowsClaude Code, Codex and Cursor
  5. Human in the loopApprovals exactly where they matter
  6. Evals and LLM as judgeMeasure what the model gets right
  7. Guardrails and safetyPrompt injection, data leaks, failure modes
  8. Shipping AI to productionDeployed, traced and handed over
02 / The format

How the two weeks work.

Learn → Practice → Capstone

Week 1
Week 2
Step 01Days 1–2

Learn

2 live sessionsConcepts, live coding and Q&A.
Tools
  • Claude Code
  • Codex
  • Cursor
  • Copilot
Step 02Days 3–6

Practice

4 daily builds + Doubt sessionEach one drills a live-session topic.
Tools
  • LangGraph
  • CrewAI
  • MCP
  • pfPromptfoo
Step 03Day 7 → Week 2

Capstone

One client, built end to endScope it on Day 7. Ship it by Day 14.
Tools
  • Docker
  • GitHub
  • +Everything from Week 1
03 / Outcomes

What you leave with.

  • 5+

    Builds on GitHub

    Guided builds plus your final capstone.

  • 10+

    JD skills practiced

    Every core requirement from real FDE job posts.

  • 10+

    Tools, hands-on

    From Claude Code to LangGraph, CrewAI, Agno and MCP.

  • 5

    Reusable templates

    Review checklist, eval suite, architecture doc, handover doc.

Don't just say you can do the job. Show what you shipped.
Two weeks · 5+ builds
04 / Who it's for

Engineers ready to take on AI projects.

You write code and work with APIs. We take it from there.

Software engineers

You ship features. Start shipping systems.

Own the client, the architecture and the go-live.

Data & backend engineers

You know the plumbing. Add the AI layer.

Pipelines and APIs are half the job. Agents and evals are the rest.

Solutions & ML engineers

You talk to clients. Now build for them.

Go from scoping a solution to deploying it yourself.

05 / Your instructor
Satvik Paramkusham, instructor of Forward Deployed Crash Course

Satvik
Paramkusham

Founder, Build Fast with AI

A generative AI consultant who has helped numerous startups integrate AI into their products, and has trained 50,000+ professionals on applied AI. Holds Bachelor's and Master's degrees from IIT Delhi.

Professionals trained
50,000+
Bachelor's + Master's
IIT Delhi
Enrollments open

The AI role that pays you to be the engineer in the client's room.

Limited seats, so every build gets reviewed.

Two-week cohortSeats open
Forward Deployed Crash Course
November cohortUpcoming
Saturday & Sunday · 7:30 PM ISTLive
5+ guided builds · 1 final capstoneHands-on
$399 USDWORTH$500 USDCost after discountOne year course access
Limited seats per cohort
06 / From the JD

Built from real job posts.

8 requirements

Each card is a line from a real FDE job description, and the course drills every one of them. Scroll sideways.

Delivery01/08
“Embed with customers, rapidly translate complex requirements into working systems.”
Coding agents02/08
“Use coding agents (Claude Code, Codex, Cursor, Copilot) as primary acceleration tools.”
Code review03/08
“Review, validate, and refactor agent-generated code with full rigor.”
Human in the loop04/08
“Build reliable pipelines with observability, audit logging, and human-in-the-loop checkpoints.”
Multi-agent05/08
“Design multi-agent systems: roles, tool access, orchestration, handoffs, state, and error recovery.”
Integrations06/08
“Build reliable, auditable integrations with enterprise systems (ERP, CRM, data warehouses).”
Reliability07/08
“Build in fault tolerance: graceful degradation, circuit breakers, retry logic, and failover.”
Enterprise safety08/08
“Test AI systems rigorously, accounting for non-deterministic outputs; document known limits for safe handover.”
07 / FAQ

Questions, answered.

What is a Forward Deployed Engineer?
A software engineer who works directly with clients. They scope the problem, build the AI system with the client's team and stay until it is live.
What does an FDE do day to day?
Turn client briefs into technical specs, build agents and integrations with coding agents, test them with evals and guardrails, and hand the system over with clear docs.
Do I need ML experience?
No. You need to write code and work with APIs. We take it from there.
Is it live or recorded?
Live. You learn with Satvik in live sessions, with doubt clearing and a capstone review. You also get one year of course access to come back to every session and build.
When does the November batch start?
The November batch is our next cohort, with live sessions on Saturday & Sunday at 7:30 PM IST. We'll share the exact start date by email.
November batch

Get a seat in the November batch.

Limited seats · Every build reviewed · 50,000+ professionals trained by Satvik

Build Fast with AIbuildfastwithai.com
Forward Deployed Crash Course$399 USD