Offline Community Connect · IDC Hyderabad

Build and Ship:
Applied AI in Action

Build a working AI app. Share the link. That's it.

Prudhvi Krovvidi · Data Scientist @ Gramener · github.com/prudhvi1709

25th Apr 2026 | 10:30 AM

Scan To Open

Workshop Slides

prudhvi1709.github.io/talks/IDC-25-04-2026/idc-workshop.html

Welcome everyone. This is not a session where you watch me build something and then clap at the end. This is a working session where you build something and ship it. My name is Prudhvi. I’m a Data Scientist at Gramener, and I spend a lot of my time building practical AI workflows and prototypes. But today is not really about my background. Today is about making sure each of you builds something real and shares a link before we leave. The bar today is not production-ready. The bar is one real link that does one useful, interesting, or funny thing. If it is ugly but it works, that still counts as a win. I want you to remember one constraint through the whole workshop: one user, one input, one output. The fastest way to fail in a two-hour session is to make the idea too ambitious. Keep these slides open. I also have backup tools ready if anything breaks. If you get stuck later, I will help you cut scope, simplify the app, and get you to a working version.

One Rule

"By the time we finish,
paste one link
in the group."

What's Possible in 2 Hours

🧪

HypoForge

Upload data. AI generates hypotheses. Stats validates them. Zero backend.

🔓

Jailbreak Arena

A game where you craft prompts to break an LLM. Built for PyCon 2026.

⚡

Built Live

Idea → working app in under 5 minutes. You just watched it happen.

These were all built by one person. Today you build yours.

One Model Is Not The Whole Story

1 What LLM do you use most today?
2 What coding agent do you reach for first?
3 When one model gets stuck, what do you switch to?
4 Which tool do you trust for coding, research, and presenting?

Your Toolkit — All Free

aipipe

Free LLM API access.
No credit card needed.

aipipe.org

Claude Artifacts

Build and run apps
directly in the browser.

claude.ai

ChatGPT

Free tier works.
Great for quick prototypes.

chatgpt.com

Gemini

Canvas mode lets you
build and iterate fast.

gemini.google.com

Nothing to install. Everything runs in your browser.

Idea Accelerators

Don't build a generic chatbot. Pick one messy input, one sharp user, one surprising output.

Use Your Own Data

Bank statement → money story
Calendar + deadlines → student cockpit
Screen time / health logs → behavior mirror

Use A Live Signal

Camera → posture or drowsiness coach
Voice / journal → dream or mood mapper
Location / meetings → timely assistant

Make It Weird Enough To Share

Text / diary → soundtrack or trailer
Group chat → roast or memory map
Unread PDF → visual explainer

Best prompts start with: "Take this input, do one smart thing, return something I can use or share."

Five Things That Actually Help

1 Plan before you prompt. Three sentences about what you're building saves 30 minutes of fixing.
2 One thing at a time. One big prompt breaks everything. Three small ones build something.
3 Read what it generates. Code that looks right and code that is right look identical. Until they don't.
4 Test before adding more. The whack-a-mole problem is real. Build 20 lines, test, then continue.
5 Imagination is now the bottleneck. Not your ability to code. Knowing what to build is the hard part.

The Build

Round 1

15 min

Pick the idea. Scope it down.

One user. One input. One output. Start smaller than you want.

Round 2

25 min

Get version one working.

Ugly is fine. Broken edges are fine. Make the core loop work.

Round 3

20 min

Add one thing that feels smart.

Use your data, a live signal, or a better output. Not all three.

Round 4

20 min

Polish, test, publish.

Fix obvious bugs. Add a short intro. Get a link you can share confidently.

Final 20 Min

Demo, debrief, Q&A.

Share links. Talk about what worked, what broke, what surprised you, and what you would do next.

This is the structure for the build window, and I want you to follow it instead of improvising your own timeline. In round one, pick the idea and cut scope. Before you prompt anything, write the app in three sentences: who it is for, what goes in, and what comes out. In round two, get version one working. Ignore polish. Ignore branding. Ignore fancy extras. Just make the core loop work. In round three, add one thing that feels smart. One better output, one better prompt, one real dataset, or one useful visualization. Not all of them. In round four, polish, test, and publish. Add a title, add one-line description, make sure it works, and get the link ready to share. If your build keeps failing, cut aggressively: remove auth, remove the database, remove extra pages, hardcode sample data if needed, and rebuild around the simplest working flow.

Look at That Group Chat

Every link in there was built today,
by someone who walked in without a plan.

That is what building with AI looks like.

github.com/prudhvi1709 All my talks kprudhvi71@gmail.com