AgenCross

Gamified Learning - Reclaim your vocabulary and cognitive power in the age of AI. Think. Learn. Grow.

Comment

Inspiration

Right now, as AI writes our emails, answers our questions, and solves our problems, we're witnessing the greatest cognitive trade-off in human history. Fine, we gained convenience, but what are we losing?, our ability to think critically, to reason deeply, and to remember without googling or vibing AI. This is making us cognitively lazy.

As privacy is what we're sacrificing for the Internet, so as cognitive ability is what we're sacrificing for AI. And unlike privacy, you feel this loss – in forgotten cognitive, to slower decision-making, and diminishing creativity.

Studies shows that our working memory, pattern recognition, and vocabulary retention are declining faster than ever, as we're outsourcing our thinking to AI, and our brain is paying the price.

The study paper on AI makes you smarter but none the wiser: The disconnect between performance and metacognition by Fernandes D., et al. found that people using LLM showed better logical-reasoning ability than those who worked without AI; however, these improvements in thinking did not lead to corresponding gains in their awareness or monitoring of their own thinking (metacognition). Building on their results, they recommend creating new interactive AI interfaces that actively support metacognitive skills, helping users better assess and track how well they are performing.

While not downplaying the importance of AI impacts, the future doesn't have to be a choice between AI convenience and human intelligence. With AgenCross, you get both.

What it does

AgenCross a word puzzling gamified learning platform, drawing from diverse knowledge and cultural themes to improve vocabulary, educational understanding, and cultural awareness, powered by AI, but designed to strengthen the very cognitive muscles AI threatens to atrophy.

Firstly, AI curates intellectual puzzles spanning history, science, culture, sports, and more - topics from diverse knowledge based, but here's the twist (HITL): humans validate every puzzle before it reaches you. It's AI-powered, human-perfected learning.

Secondly, while you're enjoying the beautifully designed, animated puzzles, AgenCross will be doing something revolutionary behind the scenes, by using Baddeley's Working Memory Model (WMM) alongside Statistical Correlation Analysis to measure five (5) cognitive domains in real-time:

  1. Verbal Comprehension – how well you process language (vocabulary)
  2. Your Working Memory – your mental workspace capacity
  3. Your Processing Speed – how fast you think
  4. Your Fluid Reasoning – your problem-solving prowess, and
  5. Your Long-Term Retrieval – your knowledge recall.

AgenCross then provides you with detailed metrics of your cognitive skills in a carefully designed graphical illustrations to be able to track your cognitive progress and performance.

Lastly, while measuring your cognitive performance, the AgenCross Learning Feature provides you with educative learning content on each words based on topic context, and you can have a Real-time Voice Conversation on the topic context - think of this as having conversation with a lecturer or a subject instructor.

How we built it

AgenCross is built on modern technology tools and infrastructures such as FastAPI, Google Cloud Platform (Cloud Run, Cloud Build, Gemini, FCM, and Custom Search), Docker, Flutter, ElevenLabs Agent, etc.

Frontend

  • Flutter: Cross-platform (Android & iOS) mobile apps with key integrations with WebSocket (real-time "Voice-to-Voice" streaming), Firebase Cloud Messaging (FCM - for receiving push notifications from the backend), etc.

Backend

  • API Framework - FastAPI (Python): The core choice for high-concurrency and native asynchronous support. With admin backend management built with HTML/CSS/JavaScript.
  • Task Orchestration - Asyncs Brokers (with Redis): Handles the heavy lifting of generating crosswords, processing learning content, push notifications, metrics, etc.
  • ORM / Database Access - SQLAlchemy: Bridges the Python code to the relational data.
  • Real-time Layer - WebSockets: Integrated within FastAPI to handle the duplex audio stream with ElevenLabs Agent.

AI/ML

  • LLMs: Gemini 2.5 Flash (speed/efficiency for structured crosswords and learning contents), Gemini 3 (ElevenLabs built-in LLM, with backup of Gemini 2.5-Flash Fallback for Voice Agent Conversation).
  • Audio/Voice - ElevenLabs: Voice Agent/ASR/TTS for voice-to-voice learning experience.
  • Search & Media: Google Custom Search and Image APIs for fetching learning images and references.

DevOps, CI/CD & Infrastructure

  • Containerization - Docker: Implied by the Container Registry.
  • Orchestration - Cloud Run: Serverless container execution that scales based on traffic.
  • CI/CD Pipeline - Cloud Build: Automatically builds, tests, and pushes images to the registry.
  • Traffic Management - Google Cloud Load Balancer: Distributes incoming traffic and handles SSL termination.

Challenges

  • Real-Time Audio Streaming: Integrating Gemini + ElevenLabs for seamless voice conversation, with memory and context management.
  • Context Enforcement: Ensuring the conversation stays on topic and gracefully handles user attempts to divert.
  • Intelligent Theme Detection: The system must automatically detect if a topic is mathematical and route requests to specialized generators.
  • Clue Solve Timing Metrics: Computing the aggregating time spent on clues due to interlocking words was challenging, as it also need to consider players back-and-forth between clues as well.

Accomplishments

AgenCross represents a significant advancement in educational technology by empowering users to monitor and improve their vocabulary and cognitive skills through an engaging, interactive words puzzling.

  • Leveraging the power of Google Gemini, AgenCross delivers rich, topic-specific learning content that enables users to deepen their understanding and explore subjects in greater detail.
  • The integration of Google Custom Search ensures that users have access to curated images and authoritative references, enhancing the learning experience with relevant visual aids and further reading materials.
  • Additionally, AgenCross utilizes ElevenLabs Conversational AI Agents to facilitate natural, voice-based dialogue, allowing users to engage in seamless audio conversations that reinforce learning and provide instant feedback.

This holistic approach not only supports vocabulary acquisition and cognitive development but also fosters curiosity and sustained engagement, making AgenCross an invaluable tool for learners seeking to expand their knowledge and skills in a dynamic, user-friendly environment.

What we learned

  • A/B testing is very important when building AI-powered platform
  • Prompting for crosswords, most importantly, when it involves auto-detection for mathematical based puzzles, the prompts needs to be carefully designed and properly tested
  • Voice-based system architecture platforms needs to be properly planned

What's next for AgenCross

  • Personalized Learning Paths: Implement adaptive algorithms to tailor content and challenges to each user’s skill level and interests.
  • Voice-Driven Gameplay: Functionality to allow user play solve puzzle clues via voice prompt to enhance the speaking and listening power, and providing smoothly and more engaging experience.
  • Self-curated Puzzle: Feature to give user the ability to be able to generate their own personalized puzzle to master their cognitive on the intended topic for either assessment preparation or professional growth.
  • Multilingual Support: Expand content and voice agents to support multiple languages for a broader audience.
  • Community Features: Enable user forums, collaborative challenges, and peer-to-peer learning.
  • Integrations: Connect with other educational institutions and tools for a richer ecosystem and more structured, tailored curriculum-based learning contents.
Share this project:

Updates