LazyLens AI：以 AI 將原始 CSV 數據轉化為可行洞見

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

What I Built

I built LazyLens AI, an AI-assisted data analysis application designed to make exploring CSV datasets easier for students, beginners, and small businesses.

Working with raw data often involves repetitive tasks such as inspecting columns, identifying missing values, calculating statistics, and creating visualizations before meaningful patterns become clear. LazyLens AI aims to simplify this workflow by bringing these essential steps into one application.

The project follows a simple approach: let deterministic data analysis handle the calculations, then use AI to help explain the results.

The planned analytics workflow includes:

• Uploading and previewing CSV datasets.
• Profiling columns and understanding dataset structure.
• Identifying missing values, duplicates, and data quality issues.
• Computing descriptive statistics using TypeScript.
• Exploring data through visualizations.
• Generating AI-assisted insights based on computed statistics.
• Extending the experience with natural-language questions and report exports.

The goal is to reduce repetitive analysis work and make data easier to understand without requiring every user to write code.

Demo

Try LazyLens AI here:

Live Demo: https://8n9xs6nlg1urdskzww7tmah8k.nativelyai.app/

Code

Explore the project repository:

GitHub Repository: https://github.com/Ayeshakhan77/LazyLens.ai

How I Built It

LazyLens AI uses a web-based architecture built around Vite, React, TypeScript, and Tailwind CSS, with Supabase planned for the backend services required by its AI features.

1. Client-side data analysis

The analytics pipeline is designed to run in the browser using pure TypeScript modules. This separates deterministic data processing from AI-generated explanations.

The intended workflow is:

CSV upload → Parsing → Data preview → Data profiling → Quality checks → Statistical analysis → Charts

This allows the core analytics experience to work independently of the AI backend.

Vitest is used to test the analysis modules and help ensure that calculations and data quality checks behave consistently.

2. AI-assisted interpretation with Gemma 2

The AI integration is designed around Gemma 2, an open-weight language model, using Groq-hosted inference.

The intended model identifier is gemma2-9b-it, subject to availability and verification during implementation.

Rather than sending raw dataset rows to the model, the application sends a compact, pre-computed statistical summary and the user's question through a Supabase Edge Function named ai-analyst.

The planned request flow is:

Frontend → Supabase Edge Function → Groq-hosted model → AI response

The GROQ_API_KEY is intended to be stored as a Supabase secret and accessed only inside the Edge Function. It must never be exposed in browser code or client-side environment variables.

The model interprets computed results rather than performing the application's numerical calculations. AI responses are designed to distinguish observed facts from hypotheses and acknowledge when the available data is insufficient.

3. Building in stages

The implementation is organized to prioritize the non-AI analytics experience first. CSV upload, parsing, data quality checks, statistics, and charts can be developed and demonstrated before connecting Supabase.

The AI features, natural-language data questions, report exports, and additional visual polish follow as subsequent development stages.

This phased approach helps keep the core application useful even while the AI integration is being configured.

Why Does Open Innovation Matter?

Open innovation matters because it gives developers more flexibility to experiment with AI, understand model behavior, and build applications around their own requirements.

For LazyLens AI, an open-weight model such as Gemma 2 offers an alternative to relying exclusively on closed model ecosystems. It allows the project to explore how an open-weight model can interpret statistical summaries and explain findings within a data analysis workflow.

The architecture also separates the model from the analytical calculations. This is important because numerical results should be reproducible and grounded in the dataset, rather than dependent on a language model's interpretation.

Using an open-weight model does not automatically make a hosted inference service private, free, or offline. The project's approach is to keep calculations deterministic, restrict the information sent for AI interpretation, and keep credentials on the server side.

My Agent Session

The project was developed with AI-assisted development as part of the implementation process.

A development session can be added here if a shareable agent session is available. This section is optional.

The Lazy Payoff

LazyLens AI aims to reduce the repetitive effort required to move from a raw CSV file to an understandable overview of the data.

Instead of manually inspecting every column, calculating basic statistics, and preparing initial charts separately, users can access these steps through one workflow.

The AI layer is intended to make the results easier to interpret by explaining computed findings in natural language.

The actual time saved will depend on the dataset, the task, and which features are available in the deployed version. I want to evaluate that improvement through real usage rather than claim an unmeasured time-saving figure.

Prize Categories

I will select any applicable partner prize categories only after confirming that the implemented project satisfies their eligibility requirements.