Engineers & Developers
Move from API calls to reliable product flows that handle context, state, and failure.
llmfn∘
From Prompt to Product
We build one product recommender and improve it layer by layer. The system evolves from search to prompts, schema, context, state, memory, tool use, and evals. Each layer reveals what the system needs next. We learn by building, breaking and refining the systems together.
No prior AI/ML experience needed. If you have written an API call and want to know what comes after the LLM call, this is for you then.
Date & time
13 June 2026
9:30 AM – 5:30 PM
Venue & Map
Price & Capacity
₹4,999 (& ~$15 OpenAI credits)
30 persons max
The How
One Product, Eight Layers.
We start with the simplest foundation: a basic search box and a ranked list. From there, we evolve the system step by step. Each additional layer is a direct response to a specific challenge. By building and refining in real-time, you will learn exactly how each component, from prompt to evals, earns its place in a production-ready product.
Every layer is motivated by a real failure. Nothing is added because it is fashionable.
Search
11 Search
Adds
Ranking logic
Improves
Users find products by keywords
Breaks
Intent is ignored
Prompt
22 Prompt
Adds
First LLM call
Improves
Understands the intent in natural language
Breaks
Output is inconsistent & unstructured
Schema
33 Schema
Adds
A strict output contract
Improves
Shape of model output is consistent
Breaks
Answer is not grounded in real product data
Context
44 Context
Adds
Relevant data injected about products
Improves
Responses are now grounded in facts
Breaks
Conversations are lost between turns
State
55 State
Adds
Conversation history retained between turns
Improves
Remembers the full conversation history
Breaks
Preferences are lost when session ends
Memory
66 Memory
Adds
User profiles written and retrieved across sessions
Improves
System builds knowledge about the user over time
Breaks
Cannot interact with the world outside the context window
Tools
77 Tools
Adds
Interaction with external systems
Improves
The system can read from and write to the real world
Breaks
Failures are invisible and hard to detect
Evals
88 Evals
Adds
A structured evaluation pipeline
Improves
Output quality is measurable and testable
Breaks
Production ready. What you build next are product decisions
One day. Four blocks.
Our day is structured around the natural evolution of an LLM-based product. We advance through four distinct blocks: laying the foundation of search and intent, engineering the structure of context and schemas, establishing conversational continuity, and finally building towards reslient quality.
Four sprints to transform a fragile prompt into a resilient product.
09:30 AM – 11:00 AM
The Foundation
Search & Prompt
- Build a baseline product recommender with search and filters.
- Watch it fail when a user asks a natural language question.
- Add a prompt layer to translate intent into data.
11:20 AM – 1:00 PM
The Structure
Schema & Context
- Define a schema contract to ensure predictable, usable data.
- Inject product specs into context to ground answers in fact.
- Connect the model to your data without losing control.
2:00 PM – 3:30 PM
The Continuity
State & Memory
- Implement session state to track context across a conversation.
- Add persistent memory to recognize returning user needs.
- Decide what data is worth keeping across sessions.
3:40 PM – 5:30 PM
The Quality
Tools & Evals
- Use tools calls to enrich and improve the generative output.
- Build an evaluation pipeline to catch errors before they ship.
- Audit the 8-layer stack against your own requirements.
The Why
The hard part starts after the first prompt.
The interesting part of LLM products is everything around the model. Not the model itself. The sequence of layers that makes it useful, reliable, and safe to ship.
This workshop teaches the sequence. It is not a checklist. It is to get you started thinking about what your product actually needs.
Each layer fixes a real failure. Skip one and the failure returns, usually at the worst possible moment, usually in front of a user.
The sequence is not obvious.
Most teams add layers in the wrong order, or add too many at once. When something breaks, they reach for a bigger model instead of the missing layer.
Each layer changes the product.
Context changes what the model can answer. State changes how it behaves across a conversation. Evals change whether you can trust it. These are product decisions, not just engineering ones.
Knowing when to stop matters.
A product that breaks badly is worse than one that does less. A product with three well-understood layers is better than one with eight poorly understood ones.
The Who
If you have written an API call and want to know what comes after.
This is for builders who want to understand how LLM products actually hold together.
“The model is only one component. The product is everything around it.”
Founders & Operators
Identify which AI features are worth building and which are distractions.
Not a Good Fit
You want a no-code workshop or a deep ML research course. This sits in the middle: practical product engineering for LLM-powered features.
Prerequisites
You should be comfortable with variables, functions, JSON, and API calls. We will use Python and JavaScript. You do not need to know both deeply, and you do not need prior AI/ML experience.
By Whom
Three people who have built it and taught it.
They've built products, taught teams, and run workshops on AI/ML, data science, and software engineering across India, Singapore, the UK, and the US.
We learn AI by building real systems under real constraints.
Amit Kapoor
Amit crafts complex data, business models, and product strategy into clear human stories. His work operates at the intersection of strategy, pedagogy, and narrative design to scale human intention.
Bargava Subramanian
Bargava brings the founder's lens to AI, making it work under real constraints. He focuses on turning prototypes into production systems by bridging business realities with AI/ML design and execution.
Anand Chitipothu
Anand designs software systems that are elegant, useful, and built to last. He focuses on developer experience and technical clarity to turn complex engineering into usable systems.
Book a Seat
Attend the LLMFn session in Bengaluru on 13th June.
Seats are ₹4,999 for a full day of in-person learning at Underline Center, with lunch and beverages included. You will also need ~$15 of OpenAI API credits for the hands-on build.
Spend a day building how modern AI products actually work.
Terms, Refunds & Privacy
Registration
Your seat is confirmed only after successful payment and receipt of the confirmation email.
Refunds & Transfers
Workshop registrations are non-refundable and non-transferable. Seats are limited and reserved specifically for each participant.
Workshop Changes
If the workshop is postponed or cancelled by the organisers, participants will be informed via email and offered an appropriate update.
Privacy
Your registration information will only be used for workshop communication, logistics, and related updates. We do not sell or share participant information with third parties.
Contact
For workshop-related questions, email us at [email protected] before registering.
FAQ
Questions before you book a seat
Practical details, prerequisites, and what to expect from the day.
Built for builders, grounded in practice
Do I need prior machine learning experience?
No. We will explain the LLM concepts you need from first principles. The workshop is designed for people with no to minimal LLM or AI/ML experience.
Do I need to know Python and JavaScript?
You should be comfortable with basic programming. We will use Python and JavaScript examples, but you do not need to be an expert in both.
Is this a prompt engineering workshop?
No. This is for product builders and goes deep into how to integrate LLMs and agents into your product.
Will we build something hands-on?
Yes. We will build one running example in layers through the day so you will see how the architecture changes as the product becomes more capable.
What should I bring?
Bring a laptop. Setup notes will include a browser, editor, Git, Python, Node.js, and any API key or provided workshop credentials.
Do I need access to an LLM app?
Yes. You should have access to at least one LLM tool such as ChatGPT, Claude, or Gemini for exploration and comparison.
Do I need OpenAI API credits?
For product integration, we will use the OpenAI API. Before the workshop, go to platform.openai.com, sign in or create an account, and add $15 of credits so your API calls work during the build.
Will this cover LangChain or specific agent frameworks?
The workshop focuses on durable concepts first: prompts, context, state, tools, evaluation, and workflow design. We will include framework-specific examples lightly when they help.