Bias-Aware Task Time Estimation and Behavioral Intervention Using Lightweight AI Models

Students routinely misjudge how long everyday academic tasks will take them. This is not a new observation — researchers have documented the tendency for decades under the label "planning fallacy" — but most tools built for students do nothing about it. A calendar app records your deadline; it doesn't tell you your estimate is probably wrong. This paper describes a system built to do exactly that. It learns from a user's own task history — how long they thought things would take versus how long they actually took — and uses that personal record to generate a more grounded prediction the next time. The methods underneath are deliberately simple: a mix of lightweight machine learning and basic statistics, chosen because they can adapt quickly even when data is sparse. When the system's prediction and the user's estimate are far enough apart, it raises a flag. The user sees the discrepancy and decides whether to revise their estimate or stick with it. That decision itself becomes part of the record. Beyond the prediction engine, the paper lays out a way to measure whether any of this is working. That means tracking estimation bias at the individual level, monitoring how predictions hold up against reality, and watching whether users who engage with the warnings actually plan more accurately over time. The focus throughout is on individual behaviour rather than group averages — partly because planning habits are deeply personal, and partly because the interesting question is whether a specific person improves, not whether a population shifts slightly. The system design and evaluation framework are what this paper covers. Testing it with real users at scale is the next step.