Bringing AI Mission Assistance to Student Space Explorers

For four decades, this nonprofit organization has honored the legacy of the Space Shuttle Challenger crew by inspiring students across the country through immersive, mission-based STEM experiences. Their network of physical learning centers and their growing Satellite Challenges program, which connects live, flight director-led missions to classrooms virtually, represents a unique intersection of human expertise and interactive education. With their 40th anniversary approaching and a strategic push to demonstrate innovation to donors, funders, and NASA-affiliated partners, the organization saw a pivotal opportunity: use generative AI not as a gimmick, but as a genuine force multiplier for the highly trained flight directors and the learners they serve.

The organization needed a working proof of concept demonstrating that a proactive, context-aware AI agent could guide students through mission interactives — offering hints without giving answers — while simultaneously surfacing real-time alerts and engagement suggestions to flight directors. The POC also needed to be compelling enough to anchor funding conversations and serve as the centerpiece of an AWS re:Invent presentation, making credibility and stakeholder-readiness as important as technical functionality.

The core challenge was not simply building a chatbot — it was proving a fundamentally different model of AI engagement. The organization's vision required an agent that acts proactively rather than reactively, initiating contact with learners based on mission state rather than waiting to be asked. That demanded genuine context awareness: the agent needed to understand where a student was in a mission phase, whether they were stuck or ahead of their peers, and what pedagogically appropriate guidance looked like — without ever surfacing a direct answer.

‍

Compounding this was the constraint of working outside the live application environment for the POC, while still producing a demonstration credible and compelling to external stakeholders. The team had to design an architecture that could tap into existing event triggers within the platform's Django and React stack without requiring deep integration, while also ensuring the AI's knowledge was appropriately scoped to mission content and authoritative sources like NASA — preventing the model from ranging beyond the carefully designed educational context.

Protagona designed a context-aware AI agent architecture built on Amazon Bedrock, using retrieval-augmented generation to ground the model's responses in the organization's mission content and curated external sources aligned with each mission theme. Rather than a generic large language model deployment, significant investment went into prompt engineering to encode the pedagogical constraints central to the organization's educational philosophy — guiding learners toward solutions through questioning and scaffolding, never by providing direct answers. This approach preserved the integrity of the learning experience while making the AI feel like a natural extension of the mission environment.

‍

The POC was scoped around the Destination Moon mission — chosen for its relevance to upcoming NASA lunar programs and its potential for donor-facing storytelling — and structured to demonstrate the full proactive engagement loop within a single mission phase. By interfacing with existing progress-tracking triggers already present in the Django and React platform rather than rebuilding instrumentation from scratch, the team delivered a demonstrable, credible proof of concept the organization could place directly in front of funders and present at AWS re:Invent. A parallel flight director alert layer was also built, giving mission staff real-time visibility into learner state across up to 30 simultaneous participants.