SocratesAI - Adaptive Socratic AI Tutor

Transforming learning from passive lectures into dynamic, interactive collaboration with an adaptive AI Socratic mentor.

Comment

Inspiration

Traditional online learning often defaults to static, one-size-fits-all lectures where students passively consume content. Even modern generative AI tools frequently dump massive text summaries rather than fostering deep, interactive comprehension. We wanted to transform online education by creating an active, collaborative AI mentor—inspired by the classical Socratic method—that adapts dynamically to a learner's exact mental model, gauges their depth in real time, and patches misconceptions on the fly.

What it does

SocratesAI is an adaptive, collaborative Socratic tutor that guides scholars, senior engineers, and students through complex technical subjects without delivering passive lectures.

  • Dynamic Diagnostic Onboarding: Evaluates prior domain knowledge and tailors lessons to configurable academic depth levels (ranging from Foundational to PhD / Advanced Research).
  • Interactive Socratic Dialogue: Delivers concise concept breakdowns paired with probing Socratic checks that test deep system invariants over simple memorization.
  • Live Knowledge Graph Patching: Dynamically modifies the student’s learning roadmap in real time, injecting remedial sub-lessons whenever misconceptions are detected.
  • Multimodal Avatar Mentor & Voice Integration: Features an interactive lip-synced SVG/Canvas avatar, real-time speech synthesis, and speech-to-text voice input.
  • Cross-Session Memory Bank: Maintains a persistent UserKnowledgeGraph in Firestore tracking mastered concepts, recurring weak spots, and learning velocity across sessions.

How we built it

  • LLM Engine: Gemini 2.5 Flash / Gemini 3.6 Flash via the official @google/genai SDK for low-latency reasoning and dynamic curriculum construction.
  • Agent Architecture: Google Agent Development Kit (ADK) standards running a structured multi-agent workflow for diagnostics, lesson generation, answer evaluation, and live graph mutation.
  • Backend Runtime: Express / FastAPI backend running in containerized environments targeting Google Cloud Run.
  • State & Memory: Firestore for persistent cross-session knowledge storage and user graph persistence.
  • Frontend & UX: React, Vite, and Tailwind CSS featuring interactive system topology diagrams, state invariant mathematical views, and canvas-rendered avatar animations.

Challenges we ran into

  • Real-Time Graph Mutation: Seamlessly patching node graphs live during active Socratic loops required carefully synchronized JSON schema enforcement to avoid breaking student progress state.
  • API Quota & Rate Limit Resiliency: Managing high-throughput multi-turn dialogues required implementing robust fallback chains across model aliases and strict control-character JSON sanitization.
  • Audio & Lip-Sync Synchronization: Aligning browser SpeechSynthesis events with procedural SVG mouth canvas animations and real-time visualizer spectrums without UI latency.

Accomplishments that we're proud of

  • Built a truly adaptive collaborative partner that doesn't just answer questions, but actively tests, guides, and evolves with the user's level of understanding.
  • Successfully implemented real-time curriculum graph patching based on live misconception detection.
  • Created a seamless multimodal learning experience featuring live voice interactions, interactive architecture topologies, and formal academic depth selectors.

What we learned

  • Structuring system prompts with strict JSON schemas is critical for enabling LLMs to reliably output state updates that drive interactive graphs.
  • True AI-human collaboration requires multimodal feedback—combining audio visualizers, voice logs, and dynamic visual diagrams makes complex learning significantly more engaging.

What's next for SocratesAI - Adaptive Socratic AI Tutor

  • Multi-Agent Debates: Introduce opposing AI peer agents to simulate academic thesis defenses and code review roundtables.
  • Expanded Multimodal Vision: Allow students to upload handwritten equations, whiteboard architecture diagrams, or live terminal logs for instant Socratic analysis.
  • Enterprise LMS Integration: Connect SocratesAI memory banks directly with institutional learning management systems and internal enterprise documentation platforms.

Built With

  • adaptive-learning
  • api
  • apps
  • cloud-hosting
  • css3
  • engineering
  • gemini-api
  • generative-ai
  • google-ai-studio
  • html5
  • interactive-ui
  • javascript
  • llm
  • next.js
  • prompt
  • react
  • serverless
  • socratic-ai
  • state-management
  • tailwind-css
  • typescript
  • vercel
  • web
Share this project:

Updates

posted an update —

The demo video has been updated to include the requested Google Cloud backend proof. The video now demonstrates the backend running live on GCP by showing the Google Cloud Console usage metrics and Cloud Run logs. The main submission form has been updated with the new video link.

Log in or sign up for Devpost to join the conversation.