DEV Community: Philip Ganchev

Why I'm Still Thinking About My Hackathon Project Weeks Later

Philip Ganchev — Tue, 22 Jul 2025 13:40:44 +0000

This is a submission for the World's Largest Hackathon Writing Challenge: After the Hack.

So I built this thing called Metric Moon for the World's Largest Hackathon. Basically, you ask your data questions in plain English and get instant, visual answers. Pretty standard AI stuff, right? I thought I was done with it after submission.

Then last week I was grabbing coffee with Maria (she runs this small animal rescue), and she's pulling her hair out over a laptop full of spreadsheets. Trying to figure out which adoption campaigns work best, what times of year they get more surrenders, basic stuff that could help them save more animals.

"I know the patterns are in here," she says, scrolling through endless rows. "But learning Excel formulas is like learning a foreign language, and we're all volunteers."

And I'm sitting there thinking... wait. This is exactly what Metric Moon does. She could just ask "which adoption events had the highest success rate?" and get an actual answer instead of spending her Saturday afternoon fighting with VLOOKUP functions.

The thing that's been bugging me

It's not that Maria isn't smart - she's brilliant at what she does and has saved hundreds of animals. But she's locked out of her own data because the tools assume you either have a CS degree or a $50k budget.

I used to think this was just "how things work." You want insights? Learn a query language/s. Can't learn it? Hire someone. Can't afford that? Too bad.

But now I'm wondering why we accept that. Like, we figured out how to make smartphones intuitive enough that anyone's grandparents can video chat with their grandkids. Why are we still gatekeeping data behind technical barriers?

What I actually learned building this

The technical stuff was honestly pretty straightforward. SkyAI Agents does most of the heavy lifting, React and Recharts handle the visuals, nothing revolutionary. I started with space mission data because it seemed like a cool demo dataset.

But what caught me off guard was realizing how many people like Maria are out there. People who have valuable data but can't do anything meaningful with it because the learning curve for data tools is just brutal.

It's not that the technology is missing; we have incredible databases, powerful analytics tools, and now sophisticated AI models. The problem is that none of it is accessible unless you already speak the technical language.

Where I'm going with this (maybe)

I keep going back to Metric Moon, not because I have some grand business plan, but because I can't stop thinking about Maria and her spreadsheets.

The space missions dataset was fine for a demo, but what if this worked with any database? What if Maria could upload her adoption data and just... ask it questions? What if my food truck friend could connect his POS system and improve his sales?
I don't know if this turns into anything bigger. Maybe I'm overthinking a weekend project. But I keep coming back to this idea that access to your own insights shouldn't require a technical degree.

The messy reality

Look, I'm not trying to put data analysts out of work or pretend AI solves everything. Complex analysis still needs experts. But asking basic questions about your own data? That shouldn't be rocket science.

The hackathon got me thinking about who actually gets to benefit from the stuff we build. Most hackathon projects die after submission, but this one keeps poking at me.
Maybe that means something. Maybe it doesn't. But I'm going to keep working on it and see what happens.

Want to see Metric Moon in action? Check out the demo and my original technical deep-dive.

From Plain English to SQL: Building Apps with AI Data Agents

Philip Ganchev — Fri, 04 Jul 2025 17:27:15 +0000

Ever found yourself bouncing between SQL IDE, BI dashboard, and spreadsheets just to answer one simple business question? Yeah, me too. That's exactly why I built Metric Moon - an app that lets you ask your data questions in plain English and get instant, visual answers.

"How many missions reached Mars each decade?"
↓ seconds later ↓
📊 Interactive bar chart + raw table + copyable SQL

The Magic Behind the Scenes: AI Data Agents

Here's what makes this possible - AI data agents. Think of them as your personal data analyst that never sleeps, never gets tired of your questions, and speaks both human and SQL fluently.

What Are AI Data Agents?

AI data agents are intelligent systems that can:

Know your data schema, table relationships, and business context
Generate optimized SQL queries that respect your database structure
Execute those queries against your database
Present results in the most meaningful way

The key difference from generic AI? These agents actually understand your data - they know which tables connect to which, what your columns mean, and how your business logic works. Instead of learning SQL syntax or figuring out which tables to join, you just ask questions like you would to a colleague.

Building Metric Moon: From Idea to Reality

I submitted Metric Moon to the Largest hackathon by Bolt, and let me tell you - AI data agents made the impossible possible in a hackathon timeline.

The Architecture

Frontend (React + TypeScript) 
    ↓
SkyAI Agent API
    ↓  
Database (Space Mission Data)
    ↓
Interactive Visualizations (Recharts)

Key Features

🗣️ Natural Language Interface
No more remembering table names or SQL syntax. Just ask:

"Show me launch trends over time"
"What's the relationship between spacecraft mass and mission cost?"
"List all currently active missions"

🤖 Smart Visualization Selection
The agent doesn't just return data - it chooses the best chart type automatically:

Mission counts by destination → Pie chart
Trends over time → Line chart
Comparisons → Bar chart

🔍 Full Transparency
Users can see the generated SQL query, because trust is built on transparency.

How SkyAI Agents Made This Possible

The real star of the show is SkyAI Agents. The best part? SkySQL provides a no-code solution for creating data agents through their UI.

Creating Your Data Agent (No Code Required!)

1. Sign up for SkySQL
Create your SkySQL account to get started with AI-powered data agents
Go to SkySQL

2. Navigate to SkyAI Agents
Access the SkyAI Agents page from your SkySQL dashboard

3. Add your SQL datasource
Connect your database or data warehouse to SkySQL

4. Create an agent
Follow the on-screen steps to create your AI data agent

That's it! Your agent now has knowledge of the specific schema, tables, and columns you selected during setup, and can intelligently answer questions about that data.

Using Your Agent in Code

Once your agent is created, integration is incredibly simple.

const askQuestion = async (question, agentId) => {
  const response = await fetch('https://api.skysql.com/copilot/v1/chat', {
    method: 'POST',
    headers: {
      'Content-Type': 'application/json',
      'X-API-Key': YOUR_API_KEY
    },
    body: JSON.stringify({
      agent_id: agentId,
      prompt: question,
      config: {} // Optional configuration object
    })
  });

  const result = await response.json();

  return {
    prompt: result.prompt,
    answer: result.response.content,
    sql: result.response.sql_text,
    columns: result.response.col_keys,
    error: result.response.error_text
  };
};

The API returns a structured response with:

content: The natural language answer
sql_text: The generated SQL query
col_keys: Column names from the query results
error_text: Any error messages if the query fails

For complete API documentation and examples, check out the SkySQL OpenAPI specification.

The Developer Experience

What amazed me most was how fast this was to build. Traditional approaches would require:

Setting up an LLM infrastructure
Building a query generation system
Implementing safety checks
Managing conversation context

With SkyAI agents, I focused on what matters - the user experience and visualization logic. The AI heavy lifting was handled by the platform.

Try It Yourself

Ready to build your own AI-powered data app? Here's how to get started:

Get your API key from the SkySQL Portal
Create a data agent with your database
Start asking questions via the API

The SkyAI Agent API docs and OpenAPI specification have everything you need to get started.

The Future is Conversational

We're entering an era where the interface between humans and data isn't charts and dashboards - it's conversation. AI data agents are making this possible today, not tomorrow.

Your users shouldn't need to learn SQL to understand their data. They should just ask questions and get answers.

What would you build with AI data agents? Drop your ideas in the comments - I'd love to hear what problems you'd solve!

Want to see Metric Moon in action? Check out the demo. Built with SkyAI Agents, React, and a lot of curiosity about what's possible when AI meets data.

How I Built an AI-Powered Food Label Scanner That Keeps You Safe

Philip Ganchev — Tue, 25 Feb 2025 23:44:07 +0000

As developers, we often create solutions to problems we personally face. The idea behind Vitalscry was born from a simple frustration: standing in grocery store aisles, squinting at tiny ingredient lists, and wondering what half those chemicals actually were—and whether they were safe to consume.

The Problem: Food Labels Are Complex

Food labels today contain dozens of ingredients, many with scientific names that might as well be in another language. For people with allergies, dietary restrictions, or those simply trying to eat healthier, every shopping trip becomes a research project.

Traditional food scanning apps rely on massive databases that need constant updating as products change. They're difficult to maintain and often lag behind the latest product formulations.

Solution: AI-Powered Ingredient Analysis

Vitalscry takes a fundamentally different approach. Instead of a traditional database, I built an AI agent that:

Scans any ingredient list through your phone's camera
Analyzes each component in real-time
Identifies potential concerns including allergens, artificial additives, and controversial ingredients
Explains unfamiliar substances in plain language

Technical Implementation: Why AI Makes Sense

Choosing to build around an AI agent rather than a traditional database offered several advantages:

Simplified Development

Building and maintaining a comprehensive ingredient database would require:

Cataloging hundreds of thousands of products
Constant updates as formulations change
Regional variations for international users

This approach eliminated these significant challenges, allowing me to focus on core functionality.

Future-Proofing

Food manufacturers introduce new ingredients regularly. With an AI-powered system, Vitalscry can analyze novel ingredients without waiting for database updates.

Contextual Understanding

Unlike simple database lookups, the AI considers ingredient combinations and concentrations, providing nuanced analysis rather than binary "good/bad" labels.

How It Works

Point your camera at any ingredients list
Wait a few seconds while the AI agent analyzes the image
Review the breakdown of potential concerns
Look at a plain-language explanation of each ingredient

The app uses a combination of OCR technology to capture the text and a custom-trained AI model to analyze each ingredient's safety profile based on scientific literature and nutritional research.

The Challenges

Accuracy vs. Speed

Finding the right balance between thorough analysis and quick response times was crucial. Users won't wait 30 seconds in a grocery aisle, but oversimplified analysis defeats the purpose.

Solution: An optimized AI model specifically for ingredient analysis, pruning unnecessary functions to maintain a 5-8 second response time while preserving accuracy.

Explaining Without Alarming

Many ingredients with chemical-sounding names are perfectly safe, while some natural-sounding ones can be problematic for certain individuals.

Solution: The explanations were designed to be educational rather than alarmist, providing context about why certain ingredients might be flagged and for whom they might be concerning.

What's Next for Vitalscry

I'm currently working on:

Personalized profiles for specific allergies and dietary preferences
Alternative product recommendations
Enabling citations

Try It Yourself

Vitalscry is available for both iOS and Android. I'd love feedback from the dev community on both the technical implementation and user experience.

Has anyone else experimented with AI agents as alternatives to traditional databases? I'd be interested to hear about your experiences in the comments!