Predictive Student Success Agent

Inspiration

Students rarely struggle overnight. Their risk usually appears through small signals such as declining marks, low attendance, incomplete assignments, and poor recent performance. The problem is that faculty cannot manually monitor these signals continuously for every student.

We wanted to build an AI agent that does more than predict academic risk. It should detect the risk, understand why it is happening, take appropriate action, and monitor whether the student improves.

This led us to build Predictive Student Success Agent.

What it does

Predictive Student Success Agent is an autonomous AI system for early identification and intervention of academic risk.

The agent:

Continuously analyzes student academic signals.
Predicts the student's risk level using a machine learning model.
Identifies the major factors contributing to the risk.
Generates a personalized intervention plan.
Recommends relevant learning resources and actions.
Notifies the student or mentor when intervention is required.
Monitors subsequent performance.
Escalates cases to a human mentor when the student does not improve.

Instead of stopping at:

Data → Prediction

our system creates a closed loop:

Detect → Understand → Act → Monitor → Escalate

The goal is not to replace teachers. It is to remove repetitive monitoring work so educators can focus on meaningful human intervention.

How we built it

We built the system around the Strands Agents SDK, using an AI agent as the central decision-making layer.

The agent can interact with dedicated tools for:

Student data retrieval
ML-based risk prediction
Risk-factor analysis
Personalized intervention generation
Learning-resource recommendation
Notifications
Progress monitoring
Human escalation

The machine learning component processes academic features such as attendance, assessment performance, assignment completion, recent scores, and previous performance.

The agent combines these predictions with contextual reasoning to determine the appropriate next action.

Our application consists of a web-based interface, backend services, machine learning components, the Strands agent, and supporting AWS services.

Challenges we ran into

The biggest challenge was making the system genuinely agentic rather than simply adding an LLM to a prediction dashboard.

We had to design clear tools and decision boundaries so that the agent could determine when to retrieve information, make a prediction, recommend an intervention, monitor the outcome, and escalate to a human.

Another challenge was balancing automation with human oversight. Academic decisions can affect students significantly, so the system is designed to support mentors rather than make irreversible decisions on their behalf.

We also focused on using anonymized or synthetic student data for our demonstration.

Accomplishments that we're proud of

We are proud to have transformed a traditional predictive analytics workflow into an action-oriented AI agent.

Our key accomplishment is the closed-loop workflow:

Predict → Explain → Intervene → Monitor → Reassess

The system does not simply tell a mentor that a student is at risk. It determines what can be done next and tracks whether that intervention is effective.

We are also proud of integrating the predictive ML component with an autonomous agent workflow using Strands.

What we learned

We learned that building an effective AI agent is not just about using a powerful language model. The quality of the system depends on how well the agent's tools, data, reasoning flow, and human-in-the-loop boundaries are designed.

We also learned how predictive models can become significantly more useful when their outputs are connected to real actions and continuous feedback.

Most importantly, we learned to design AI around a real human problem rather than building technology for its own sake.

What's next for Predictive Student Success Agent

Our next step is to expand the system from academic-risk monitoring into a broader student success platform.

Future versions could integrate learning management systems, attendance systems, assessment platforms, and institutional communication channels.

We also plan to improve personalized interventions through long-term student behavior patterns, strengthen model explainability, and provide mentors with actionable cohort-level insights.

Our vision is simple:

Identify students who need help earlier, provide the right support faster, and keep humans in control of important decisions.