DEV Community: Srija Bhattacharya

PitchPilot AI: Building an AI Stadium Assistant for the FIFA World Cup 2026 ⚽🤖

Srija Bhattacharya — Sun, 19 Jul 2026 00:46:21 +0000

Building PitchPilot AI in Public 🚀

A few nights ago, I found myself wondering about something that most people probably don't think about until they're actually inside a stadium.

Imagine you've travelled to another country for the FIFA World Cup. The stadium is packed with tens of thousands of fans. You're trying to find your gate before kickoff, your phone has no signal, and you don't understand the local language.

Suddenly, something as simple as finding your seat becomes stressful.

That question stayed with me:

What if AI could make the match-day experience just a little easier?

So I started building PitchPilot AI—a GenAI-powered stadium assistant designed for the FIFA World Cup 2026.

The goal wasn't to build another chatbot. I wanted to build something that could actually solve problems fans and stadium staff might face during a match.

🛑 The Problem

Modern stadiums are incredible, but they can also be overwhelming.

Fans often struggle with:

Finding the correct entrance or seating section.
Understanding stadium rules and bag policies.
Communicating with staff because of language barriers.
Getting quick information when the venue is crowded.

For stadium staff, handling thousands of visitors while responding quickly to incidents is equally challenging.

I wanted one application that could help both.

💡 The Solution

PitchPilot AI combines AI with practical stadium utilities instead of relying on AI for everything.

Rather than making users search through menus or FAQs, the assistant lets them simply ask questions naturally and get the information they need.

⚽ Features

🗺️ Smart Stadium Navigation

Users can find the quickest route to their seats using an interactive stadium map.

The navigation considers simulated crowd conditions, helping fans avoid congested areas and reach their section more efficiently.

🤖 AI Stadium Assistant

Which gate should I enter?
Can I carry a backpack?
Where are the food courts?
Is wheelchair assistance available?
How do I reach the parking area?

Instead of searching through multiple pages, users simply ask.

🌍 Multilingual Support

Since the World Cup brings together fans from around the globe, PitchPilot AI supports conversations in six languages, making communication much easier for both visitors and stadium staff.

🚨 Incident Response Assistant

I also wanted the app to be useful behind the scenes.

PitchPilot AI can generate structured incident-response checklists within seconds, giving staff a quick starting point during emergencies.

⏳ Live Match Experience

Before kickoff, the dashboard displays a live countdown.

Once the match begins, the interface automatically switches to a live match simulation, making the experience feel much more engaging.

✅ Testing & Validation

Building the features was only part of the process—I also wanted to make sure they behaved consistently across different scenarios.

Rather than focusing only on the "happy path," I manually tested the core user journeys multiple times, including navigation, AI interactions, multilingual responses, and the live match experience.

Here's a quick overview of what I verified:

Feature	Validation
Stadium Navigation	✅ Verified route guidance between stadium sections
AI Stadium Assistant	✅ Tested common fan questions and responses
Multilingual Support	✅ Confirmed responses across supported languages
Incident Response	✅ Verified structured checklist generation
Match Countdown	✅ Confirmed automatic transition to live match simulation
Fallback Mode	✅ Tested application behavior when AI responses were unavailable

I also experimented with invalid prompts, empty inputs, and temporary AI service interruptions to ensure the application continued providing a smooth experience whenever possible.

## 🏗️ Architecture

User
      │
      ▼
HTML • CSS • JavaScript
      │
      ▼
FastAPI Backend
      │
      ▼
Google Gemini API
      │
      ▼
AI Response

The frontend handles the user interface and interactions, while FastAPI manages communication with the Gemini API. Keeping these responsibilities separate made the application easier to organize and maintain as new features were added.

## 🔒 Reliability & Error Handling

Since AI services can occasionally become unavailable, I wanted PitchPilot AI to remain useful even when AI-powered responses couldn't be generated.

To achieve this, I implemented a graceful fallback mechanism. If an AI request fails or exceeds the available quota, the application continues providing access to core non-AI features such as stadium navigation, the match countdown, and the overall interface instead of displaying a broken or unusable page.

This approach improves reliability and ensures users can continue using essential functionality even during temporary service interruptions.

## ⚡ Performance & Efficiency

Since stadium visitors often need information quickly, I tried to keep the application lightweight and responsive.

Some of the decisions that helped include:

Keeping the frontend lightweight with HTML, CSS, and JavaScript
Using FastAPI for efficient backend request handling
Calling the Gemini API only when AI assistance is actually needed
Separating AI-powered features from non-AI utilities so the application remains useful even during service interruptions

These choices helped reduce unnecessary processing while keeping interactions fast and responsive.

♿ Accessibility

PitchPilot AI is designed to be easy to use for as many people as possible.

Some of the design considerations include:

Responsive layouts for desktop and mobile devices
Clear typography for improved readability
High-contrast interface elements
Simple navigation with minimal user interactions
Straightforward conversational interface that reduces the need to search through menus

There is still room for improvement, and future versions will focus on expanding accessibility support even further.

🛠️ The Unexpected Challenge

One challenge I didn't anticipate was working within the Gemini API quota during development. While testing different prompts and refining the assistant, I occasionally reached the request limit, which interrupted my testing workflow.

Rather than treating it as just another error, it made me think about how users would experience the application if AI became temporarily unavailable.

That experience changed the way I approached the project. Instead of relying entirely on AI, I focused on designing an application that remains useful even when AI services aren't available.

It reminded me that reliability is just as important as intelligence when building real-world AI applications.

🏗️ Tech Stack

Frontend

HTML5
CSS3
JavaScript (ES6)

Backend

FastAPI
Python

AI

Google Gemini 2.0 Flash

Deployment

Docker
Vercel

Development

Git & GitHub
Visual Studio Code

I could have added more services and frameworks, but I wanted to spend my time solving user problems instead of making the architecture more complicated than it needed to be.

💭 What I Learned

Every project teaches something different.

For me, PitchPilot AI reinforced three things:

AI works best when it solves a specific problem instead of trying to do everything.
Small details—like graceful fallbacks and clear navigation—often matter more than flashy features.
Building something people can actually use is far more satisfying than simply experimenting with new technology.

🚀 What's Next?

There are plenty of ideas I'd love to explore in future versions:

Live crowd-density visualization
Indoor navigation with real-time positioning
Voice-based conversations
Public transport integration
Personalized recommendations based on ticket location
Live stadium notifications

🚀 Try PitchPilot AI

👉 Live Demo:

https://pitch-pilot-ai-beta.vercel.app/

If you have suggestions or spot something that could be improved, I'd genuinely love to hear your feedback.

Final Thoughts

PitchPilot AI started with a simple question and slowly evolved into one of my favorite projects to build.

It reminded me that creating AI applications isn't just about integrating large language models. It's about designing experiences that remain useful, reliable, and intuitive even when things don't go exactly as planned.

There's still a long list of ideas I'd like to explore, but I'm happy with how far this project has come—and even more excited about where it could go next.

Thanks for reading! ⚽🤖

Built with ❤️ by Srija Bhattacharya for Google for Developers × Hack2Skill – Build with AI | PromptWars Virtual Challenge 4

#BuildWithAI #PromptWars #GoogleGemini #WorldCup2026 #BuildInPublic

If you were attending the FIFA World Cup, what AI feature would make your stadium experience better? ⚽

🏆 How I Built Srija AI: An AI Tutor That Makes the Stock Market Easy for Beginners

Srija Bhattacharya — Mon, 13 Jul 2026 23:26:51 +0000

🎓 I Built an AI Tutor That Makes the Stock Market Less Intimidating for Beginners

#ai #python #streamlit #googlegemini #education

A few weeks ago, I was talking to a friend who wanted to learn about the stock market.

Within five minutes, the conversation was filled with words like shares, dividends, IPOs, and diversification. The response was exactly what I expected:

"This is way too confusing."

And honestly, I couldn't blame them.

Most beginner resources either throw too much information at you or assume you already know the basics. I wanted to see if AI could make that first learning experience a little less intimidating.

That's how Srija AI came to life.

The Idea 💡

Instead of building another chatbot that simply answers questions, I wanted to create something that teaches.

The goal was simple:

Explain concepts in plain English.
Use relatable examples instead of textbook definitions.
Ask questions to keep learners engaged.
Make learning feel like a conversation, not a lecture.

If a 15-year-old could understand a topic after chatting with the app, I'd consider it a success.

What Srija AI Can Do 🚀

Here's what the app currently offers:

Beginner-Friendly Lessons 📖

Instead of overwhelming users with technical jargon, the AI explains concepts using simple language.

Real-Life Examples 🌍

Concepts become much easier to remember when they're connected to everyday situations.

For example, instead of giving a formal definition of diversification, the AI compares it to carrying different snacks in your bag instead of relying on just one. It's a simple analogy, but it helps the idea stick.

Interactive Quizzes 📝

After every lesson, users can test themselves with multiple-choice questions and receive instant feedback explaining why an answer is correct or incorrect.

Ask Anything 💬

Users aren't limited to predefined lessons. They can ask follow-up questions naturally, making the experience feel more like learning with a tutor than using a search engine.

The Hardest Part Wasn't the Code 🧠

When I started this project, I assumed most of my time would go into writing Python.

I was wrong.

The real challenge was designing prompts that consistently produced beginner-friendly explanations.

My first few attempts weren't great.

Sometimes the AI gave answers that sounded like they came straight from a finance textbook. Other times, the explanations were too long or used terms that beginners wouldn't understand.

I ended up rewriting the prompts several times—changing the wording, adding constraints, testing different teaching styles, and refining the outputs until the responses felt natural.

That process taught me something I hadn't expected:

Good AI applications aren't just built with code—they're built with good prompts.

Things Didn't Always Work 😅

Like every project, this one had its share of frustrating moments.

Some days were spent building features.

Other days were spent wondering why nothing worked.

I ran into API configuration issues, deployment errors, Streamlit problems, and a few ngrok headaches along the way.

There were moments when fixing one bug seemed to create two more.

Looking back, though, those debugging sessions taught me more than any tutorial could.

Tech Stack 🛠️

I deliberately kept the stack simple:

Python
Streamlit
Google Gemini API
Google Colab
ChatGPT
ngrok
GitHub

Nothing overly complex—just the right tools to turn an idea into a working application.

What I Learned 📚

This project changed the way I think about AI.

Before building Srija AI, I thought the hardest part would be integrating an LLM.

Now I think the real challenge is understanding the person on the other side of the screen.

A technically correct explanation isn't always the most helpful one.

Sometimes a short explanation, a relatable analogy, or a simple quiz can make a much bigger difference.

That was probably my biggest takeaway from this project.

What's Next? 🚀

There are still plenty of ideas I'd love to explore:

🎙️ Voice-based tutoring
📊 Interactive stock market simulations
🌍 Support for multiple languages
📈 Personalized learning paths based on quiz performance

One step at a time.

Before I Wrap Up ❤️

I'd like to thank Dr. Pallavi Khanna Ma'am, Infosys Springboard, Skillsoft, and the AI EMPOW(H)ER Program for encouraging learners from diverse backgrounds to explore Generative AI through hands-on projects.

Coming from a commerce background, building an AI application felt like a big step outside my comfort zone. This experience reminded me that you don't need to have all the answers before you begin—you just need the willingness to learn, experiment, and keep going.

Final Thoughts

Srija AI started as a simple idea: make stock market learning easier for beginners.

Somewhere along the way, it became much more than that.

It taught me patience while debugging, the importance of prompt engineering, and that building something useful is often more rewarding than building something complicated.

If this project helps even one beginner feel a little more confident about learning the stock market, I'd say it was worth every late night spent working on it.

Thanks for reading! 😊

Built with ❤️ by Srija Bhattacharya for the #AI EMPOW(H)ER Program.

Try It Out!

🔗 Live App: https://ai-learning-buddy-srija.streamlit.app/

Would love your feedback — drop a comment below! 👇

I'd love to know—if you were building an AI tutor, what feature would you add first?

VoteSmart: Turning Election Complexity into a Conversational AI Experience!

Srija Bhattacharya — Sat, 02 May 2026 16:15:42 +0000

Building VoteSmart in Public (with Antigravity) 🚀

Elections are one of the most important pillars of any democracy—yet for many people, especially first-time voters, the process feels confusing, fragmented, and intimidating. What if understanding elections felt less like reading a dry rulebook and more like having a conversation? 💬

That question led me to build VoteSmart—an AI-powered, interactive platform designed to simplify the election process and make citizens more informed, confident, and engaged.

🛑 The Problem: Information Exists, But It’s Not Accessible

Election-related information is everywhere—government websites, news articles, PDFs—but it’s rarely:

🎀 Personalized to the user's specific context.

🌸 Interactive enough to keep users engaged.

✨ Easy to explore without feeling overwhelmed.

Most people don’t need more information; they need better ways to understand and digest it.

💡 The Idea: Make Learning Elections Conversational

VoteSmart approaches this problem differently. Instead of forcing users to search, scroll, and decode complex legal content, it creates a guided, interactive experience where users can:

🐣 Ask questions naturally and get instant answers.

🌈 Explore timelines visually to see the big picture.

🎭 Simulate scenarios to prepare for real-world hurdles.

📝 Test their understanding through interactive, fun tools.

The goal is simple: Make every user walk in confused and walk out informed.

🛠️ What VoteSmart Actually Does

At its core, VoteSmart combines AI, structured data, and a clean UI to deliver multiple learning modes.

🧠 AI Chat Assistant

Users can ask election-related questions and receive context-aware, role-specific answers powered by Google Gemini.
This assistant adapts its responses based on whether the user is a:

🍭 First-time voter looking for the basics.

🗳️ Candidate needing nomination details.

🗞️ Journalist or Student researching the process.

🗓️ Interactive Timeline

Instead of reading static lists, users can explore the election process step-by-step through:

⭐ Animated milestones showing each stage of the cycle.

🔍 Expandable details for deeper context on every step.

📢 Contextual explanations of constituency delimitation and results.

🧩 Scenario Simulator

One of the most practical features allows users to explore “what if” situations, such as:

❓ What happens if your name is missing from the voter list?

📝 What should a candidate do during the nomination filing?

🏆 Gamified Learning

To ensure civic education sticks, VoteSmart includes:

🎈 Topic-based quizzes to reinforce knowledge.

🃏 Progressive scoring and digital flashcards.

🏅 Badge rewards to celebrate milestones and progress.

🏗️ Architecture: Simple, Fast, and Scalable

I prioritized a lightweight, practical architecture to ensure a seamless user experience:

🎨 Frontend: Vanilla JS — zero unnecessary bloat for a snappy UI.

⚡ Backend: FastAPI — clean APIs and asynchronous support.

🧠 AI Layer: Google Gemini API — state-of-the-art NLP for bilingual support.

🌙 UX: Full Dark Mode support and a bilingual interface (English + Hindi).

Why I Didn’t Overcomplicate the Stack

It’s easy to fall into the trap of using a complex microservices architecture, but VoteSmart proves that you don’t need an overengineered stack to build a powerful product—you need clarity. The app is fully containerized via Docker and deployed using Google Cloud Run, keeping infrastructure simple and cost-efficient.

📝 Key Takeaways

🧸 UX Matters More Than Features: You can have powerful functionality, but if it’s not intuitive, users won’t engage.

🌱 Start Simple, Then Scale: Get the core system working reliably first before adding heavy automation.

🤖 AI Is a Tool, Not the Product: The real value comes from how you structure information to guide the user.

🚀 Practical > Perfect: A working system with clear value beats a “perfect” system that never ships.

🔮 What’s Next

VoteSmart is just the starting point. Future improvements could include:

🌍 Support for more regional Indian languages.

📊 Real-time election data integration.

🎤 Voice-based interactions for greater accessibility.

Elections shouldn’t feel like a puzzle. With the right combination of AI and design, we can make civic education engaging, understandable, and actually useful.

Built with ❤️ by Srija Bhattacharya for the #BuildwithAI #PromptWarsVirtual challenge.

Try It Out!

🔗 Live App: https://votesmart-898928325653.us-central1.run.app

Would love your feedback — drop a comment below! 👇

Given how complex elections can be, which specific part of the voting process do you think would benefit most from an AI-powered conversational guide?