Forward Deployed Engineer Program: from prompt to production.
6 weekends, one AI SaaS product, built and run in production. Learn everything on the way, from your first prompt to a model you fine-tune and host yourself.
Starts
1 Dec 2026
Duration
6 weekends
Classes
Sat and Sun, 6:00–10:30 PM IST
Batch size
30 students
Overview
Launch your own AI product
AI moves too fast to learn slowly: build one real product with AI, from first prompt to production, and learn everything on the way.
This is an AI-native cohort, not a manual coding course. Students use AI for writing code, debugging, infrastructure and deployment, the way real teams build today. What we teach is what AI can't do for them: deciding what to build, directing the AI, reviewing what it writes, and owning the result in production. Every student must be able to explain anything they ship, and Demo Week tests this.
Every student builds one product over 4 weekends: an AI customer support agent, taken from the first LLM call to a production deployment running on their own fine-tuned model. Each student picks a business variant, so every demo is different while the code and classes stay the same.
What you will learn
From prototype to production
The technical depth and engineering judgment to take AI systems all the way to production.
LLMs and prompting for production
How LLMs work, the model landscape and what it costs. System prompts, structured output, streaming, tool calling and memory.
RAG and long-term memory
Ingestion, chunking, embeddings, pgvector, hybrid search and reranking. Grounded answers with citations, and memory that lasts across sessions.
MCP, Agent Skills and orchestration
Build and deploy an MCP server, write Agent Skills, and route work between tools, documents, models and humans.
Agents and context engineering
Agent loops, stateful agents with human approval, context budgets, prompt caching and multi-agent systems.
Evals, safety and harness
Golden sets from real logs, LLM-as-judge, evals in CI, guardrails, prompt injection defence and Jev safety checks.
Fine-tuning your own model
How training works, then LoRA and QLoRA fine-tuning on your own product's conversation data.
Self-hosting on AWS
Serve your model with vLLM or Ollama on a GPU instance, with API fallback and model comparison on the same evals.
Production, security and responsible AI
CI/CD, HTTPS, autoscaling, monitoring, IAM and secrets, PII redaction, and an AI usage policy for your product.
Hands-on
What you will build
One product, shipped four times. Each week ends with a new live version on AWS.
Week 1
v0 live at a public URL
Week 2
v1 answers from real documents and remembers users
Week 3
v2 acts safely as an agent
Week 4
v3 on its own model, public demo
Full curriculum
Course curriculum
Week 0 warm-up and 4 build weeks. Open a week for its Saturday and Sunday lessons, assignment and tools.
Mentors
Learn from engineers who ship
Vijay Deore
Co-founder, AIxVeda
Builds AI systems for real businesses.
Harish Cypher
Co-founder & CTO, Dhumi
Ships AI products in production.
Nachiketh Sharma
CEO, Dhumi
Leads live builds and mentor reviews.
Sanmati Ukhalkar
CTO, Syntine
AI infrastructure for startups.
How it runs
A weekly rhythm that keeps you shipping
Saturday
Live class, Day 1, 6:00–10:30 PM IST
Sunday
Live class, Day 2, 6:00–10:30 PM IST; assignment released
Tuesday
Office hours 1: build help and debugging, 8:00–10:00 PM IST
Thursday
Office hours 2: architecture and code review with practitioners, 8:00–10:00 PM IST
Friday
Assignment due by 11:59 PM IST
Sat–Mon
Mentors score and send feedback; leaderboard updated
Assignments, scored out of 10
| Works as asked | 4 |
| Explained clearly | 3 |
| Extra effort or stretch goal | 2 |
| Submitted on time | 1 |
Certificate rules
- 1At least 80% class attendance
- 2At least 3 of the 4 weekly assignments submitted
- 3v3 live and presented in Demo Week
Why this program
Built to make you job-ready
One product, not ten mini projects
Every student launches one production-grade AI SaaS product, live with real users by the end.
AI-native, not manual coding
You use Claude Code or Cursor throughout. We teach what AI can't do for you: deciding what to build, directing it, reviewing it and owning it in production.
Practitioner mentors
Office hours twice a week are run by people who build AI products in production.
Accountability built in
Weekly scores out of 10, a leaderboard and peer review of 2 classmates' projects every week.
A real engineering setup
You work on your own AWS account with your own LLM API key, the way engineers do on the job.
Demo Week with industry judges
Mentor demos on Saturday, then a public demo day streamed on YouTube with industry judges.
Tech stack
Skills you will acquire
AI models and coding agents
- Claude API
- Claude Code
- Cursor
- Prompt engineering
- Tool calling
- Jev decision model
Agents and orchestration
- LangChain
- LangGraph
- MCP
- Agent Skills
- Context engineering
- Multi-agent systems
RAG, memory and data
- RAG
- Hybrid search and reranking
- Long-term memory
- pgvector
- PostgreSQL
- Supabase
Build and ship
- Python
- FastAPI
- Next.js
- React
- Node.js
- Docker
- Git
- GitHub
- GitHub Actions CI/CD
Cloud and production
- AWS EC2, ECR, S3, Amplify, IAM
- AWS CloudWatch and Secrets Manager
- Langfuse tracing
- Ragas and evals
Safety and responsible AI
- Guardrails and prompt injection defence
- PII redaction
- Responsible AI
Fine-tune and self-host
- PyTorch
- LoRA and QLoRA
- Hugging Face
- Unsloth
- vLLM
- Ollama
Got questions?
Frequently asked questions
Student success
What our students say
Reviews from students of our last cohort.
Ready to go from prompt to production?
Starts 1 Dec 2026 · 30 seats · ₹999 per student