Inspiration: Escaping the Digital Whirlpool
The inspiration for Aura came from a personal struggle with "cognitive drift." I found myself trapped in a late-night cycle of mindless scrolling—knowing I should sleep, yet lacking the willpower to stop. I also noticed that during deep work, unexpected interruptions triggered an instinctive irritability that didn't align with the "ideal version" of myself—the calm, professional leader I aspired to be.
I realized the problem wasn't a lack of desire to change, but a perception gap. Humans have a weak real-time perception of their own cognitive state. I needed an "External Metacognition" to help me see what I couldn't see in the moment.
What it does: The Metacognition Loop
Aura assists human self-evolution through a four-stage logic cycle:
- Contextual Observation: Ingests behavioral logs to identify drift patterns.
- Dual-Persona Profiling: Generates a Real-World Logic profile (deconstructing defense mechanisms) and a Heroic Narrative profile (motivational growth stories).
- Intervention Contracting: Users negotiate "contracts" during rational moments to authorize AI intervention during impulsive moments.
- Strategy Discovery: Analyzes failed interventions via deep-dive brainstorming to uncover blind spots and refine the strategy.
How we built it
I leveraged Gemini 1.5 Pro in Google AI Studio to move beyond simple chat and into "Reasoning Orchestration": Logic-First Architecture: Prioritized complex System Instructions to handle deep psychological deconstruction. 1M Context Ledger: Utilized the massive context window as a long-term memory for tracking persona alignment over weeks of data. Multimodal Blueprint: Designed for future Spatial-Temporal reasoning to trigger interventions via real-time video/audio sensing.
Challenges we ran into
- Bridging the Execution Gap: As a Product Manager with no prior web coding experience, the steepest challenge was transitioning from a psychological framework to a functional technical prototype in just 1.5 days.
- Decomposing System Tasks: Learning how to break down complex system architectures into executable tasks for AI was a process of trial and error. Navigating the boundary between "system prompt logic" and "hard-coded execution" required a significant shift in my mental model of development.
Accomplishments that we're proud of
- From Vision to PoC: Successfully validated a high-fidelity "Metacognition Logic Engine" that can actually distinguish between "Real-World Logic" and "Heroic Narrative" without losing consistency.
- Proactive Multimodal Intervention: Successfully validated the efficacy of AI multimodal recognition in scenarios where users are unable to actively initiate requests—such as reactive stress loops—by proving the feasibility of real-time, sensory-driven intervention triggers.
What we learned
- Prompts as Firmware: I discovered that in the era of Gemini 3, a Prompt is no longer just an instruction—it is the Logical Firmware of the application.
- Task Decomposition Methodology: I learned a new methodology for AI-native product development: focusing on "Orchestrating Reasoning" rather than just "Writing Features."
- The Boundary of AI Sovereignty: Through extensive testing in AI Studio, I gained a deep understanding of where AI's autonomy ends.
What's next for Aura - Redefining Human Agency
Current implementation is a high-fidelity Logic PoC. The multimodal real-time recognition is our primary roadmap focus, leveraging the conceptual frameworks demonstrated in the video.
Log in or sign up for Devpost to join the conversation.