Web App2025

KeplerLab Agentic

An autonomous AI assistant that executes code securely and turns files into custom study resources.

KeplerLab Agentic is an intelligent study assistant powered by autonomous AI. You can upload files or type a topic, and the AI automatically finds learning resources, writes and runs code in a secure sandboxed environment, and generates custom study tools like interactive quizzes, flashcards, mindmaps, and even multi-voice audio podcasts. It features a Next.js frontend, a FastAPI backend, and utilizes Docker for safe code execution.

The Challenge

Problem Statement

Information Overload: Finding high-quality, relevant study materials across search engines and YouTube takes too much time, and results are often filled with irrelevant spam. No Hands-on Practice: Coding students or data students cannot run example code directly inside their study tools, requiring them to switch windows and manually set up local programming environments. Passive Reading: Standard PDF readers or study documents do not support active learning methods like interactive testing, interactive mindmaps, audio podcast conversions, or real-time question answering. Security Risks: Running AI-generated code snippets on a local machine to test them can expose the developer's system to security vulnerabilities

Key Features

Autonomous AI Agent (ReAct Loop): Uses a modular "Reasoning and Acting" loop with SSE (Server-Sent Events) streaming. The agent plans, calls tools, and self-corrects in real-time.

Self-Extending Agent Tools: The AI agent has the ability to write, test, and install new tools for itself dynamically during execution.

Isolated Multi-Language Sandbox: Run code safely inside Docker containers. Supports Python, JavaScript, TypeScript, C/C++, Java, Go, Rust, and Bash with automatic timeout controls.

Smart Output & Artifact Capture: Automatically detects files generated by sandboxed code (like Matplotlib charts, Excel sheets, and images) and saves them as downloadable project artifacts.

Autonomous Web Browser: The agent can launch a headless browser (Playwright) to visit websites, scroll, click buttons, take screenshots, and extract text.

AI Resource Discovery Pipeline: Input a study topic, and the system uses SearXNG search to plan queries, filter out spam/unsafe links, and score result relevance using an LLM.

Multi-Voice Audio Podcast Studio: Convert files or notes into two-person audio discussions (host and guest) using neural text-to-speech.

Interactive Podcast Q&A: Pause the podcast at any moment to ask the host a doubt. The AI host answers the question using database info and speaks the answer back.

Study Material Generator: Instantly generate structured flashcards, graded quizzes, visual mindmaps, and even PowerPoint slide presentations from your notes.

Robust Document Processing (RAG): Upload PDFs, notes, YouTube links, or web pages. The backend automatically extracts text (OCR if needed), chunks the content, embeds it, and saves it in a vector store for semantic search.

Dynamic LLM Switcher: Switch active AI models (Gemini, Claude, GPT, Ollama, Groq, NVIDIA) and manage API keys directly from the settings panel.

Project Gallery

Take a closer look at screenshots and live visuals of the build.

Zoom Screen

Zoom Screen

Zoom Screen

Zoom Screen

Zoom Screen

Zoom Screen