DriftGuard AI

Motivation is unreliable. Systems are absolute.

Comment

Inspiration

Motivation is unreliable. Systems are absolute. Most productivity tools suffer from a fundamental flaw: they are passive. They rely on the user’s fleeting desire to succeed, acting as mere mirrors that reflect failure without resisting it. We were inspired by the concept of the Intention-Action Gap—the psychological space where ambitious goals go to die. We realized that humans don't need more "trackers"; they need accountability and a partner. We built DriftGuard to be the digital defense against goal abandonment, ensuring that personal ambitions are treated with the same non-negotiable rigor as a professional mandate.

Theme Connection

DriftGuard AI primarily aligns with TRACK 2: MACHINE LEARNING / AI.

While the project has significant overlap with Social Good by solving the universal problem of goal abandonment, its core identity is a Machine Learning/AI solution. Unlike traditional productivity tools that act as static databases, DriftGuard uses AI as a central engine for action.

We chose this track because the project demonstrates a high-level application of Agentic AI:

  • Autonomous Decision Making: The system doesn't just display data; it uses GPT-4o to analyze the "Intention-Action Gap" and autonomously determines the severity and style of intervention required.
  • Agentic Orchestration: We utilized FastAPI to link AI reasoning directly to real-world execution (Mailgun), moving AI out of the "chatbot" box and into a functional, end-to-end system.

DriftGuard represents the future of AI: a machine that doesn't just suggest, but acts to help users achieve their most ambitious goals.

What it does

DriftGuard is an end-to-end accountability ecosystem that bridges the gap between setting a goal and actually finishing it. It functions through two core agentic interventions:

  • The Architect (Agentic AI Chat): Before a goal is even saved, an AI consultant interrogates the user's intent. It forces vague ambitions into the SMART framework (Specific, Measurable, Achievable, Relevant, Time-bound), ensuring only high-clarity, executable targets reach the database.
  • The Guardian (Agentic Email Accountability): Once a goal is set, Charlie, our autonomous agent, takes over. Charlie monitors the status of goals in real-time. If a user falls behind, Charlie executes a tiered intervention strategy: from psychological nudges to formal Performance Improvement Plans (PIPs) and, ultimately, social escalation to the user’s mentors.

How we built it

We engineered a robust orchestration layer designed for real-time accountability:

  • Backend Orchestration: Built with FastAPI, which serves as the central hub orchestrating all database interactions, the agentic AI chat sessions, and the email delivery logic.
  • Agentic Reasoning: Powered by OpenAI GPT-4o in JSON mode. This allows Charlie to analyze user progress and autonomously generate contextually aware interventions based on behavioral psychology.
  • Database & Persistence: Supabase acts as our source of truth, storing user states and an agent_memory table that gives Charlie a persistent history of user behavior and past interventions.
  • Action Layer: Integrated Mailgun API to cross the digital-physical gap, delivering professionally styled "Intervention Documents" directly to the user's inbox to ensure the consequences of drift are felt.

Challenges we ran into

The most significant hurdle was API Orchestration and Authentication. Navigating the strict requirements of professional email delivery while maintaining a live, reactive backend was a massive challenge that required building robust error-shielding logic. We also struggled with Context Clarity—ensuring the AI understood exactly how a user was performing relative to their specific deadlines. We solved this by creating a dedicated enrichment layer in the backend that translates raw data into clear, actionable context for the AI before any intervention is generated.

Accomplishments that we're proud of

We are incredibly proud of Closing the Loop. It is one thing to display data; it is another to make that data act. Seeing a backend state change trigger a beautifully formatted, psychologically-tuned PIP in a real-time inbox was our "magic moment." We successfully transformed an LLM from a simple chatbot into a Systemic Agent that enforces accountability autonomously.

What we learned

The greatest takeaway from this project was that Clarity is the supreme variable. In the age of AI-assisted development, the bottleneck is no longer syntax; it is Specification. To build an agentic system that doesn't hallucinate, we had to become "Intent Architects." We learned that if the developer's logic and the system's instructions aren't perfectly clear, the AI’s execution cannot be effective.

What's next for DriftGuard AI

We are currently refining the product to move from a hackathon MVP to a full-scale startup. Our roadmap includes More psychology and interactivity with email replies to increase goal adherence and expanding "Charlie" into a multi-platform agent with integrations for Slack, Discord, and SMS to ensure there is nowhere for "silent abandonment" to hide.

Built With

Share this project:

Updates