1. What inspired us to build this? As educators and academic researchers, we spent years watching a painful paradox unfold in higher education and STEM classrooms: the most valuable learning happens through authentic, open-ended problem-solving—yet almost all classroom tech relies on shallow, multiple-choice trivia. Tools like Kahoot and Quizizz made classrooms lively, but reduced complex concepts to speed-based buzzer races. When we assigned real-world scenarios—asking students to write code, synthesize multi-step reasoning, or analyze diagrams—the grading backlog became overwhelming. We were inspired to build Saval to prove that educators do not have to choose between academic rigor and classroom excitement. We wanted a platform where authentic Problem-Based Learning (PBL) could be both deeply engaging and effortless to assess. 2. What problem were we trying to solve? We set out to solve the "authenticity vs. scale" bottleneck in formative assessment: The Grading Backlog: Qualitative assessment—evaluating student logic, running code, and marking open-ended explanations—takes hundreds of instructor hours each semester. As class sizes grow, educators are forced back into rigid multiple-choice tests. Binary, Demotivating Feedback: Traditional auto-graders mark answers strictly right or wrong. Real-world learning requires proportional partial credit and immediate, actionable guidance so students can iterate on misconceptions right away. Passive Engagement: High-stakes problem sets often cause anxiety and disengagement, while superficial gamification trivializes academic depth. Saval solves this by pairing human-calibrated multi-LLM grading (Google Gemini and OpenAI GPT) with multimodal submission capabilities (text, executable code across 15 containerized runtimes, and visual proofs), wrapped inside collaborative, team-based game modes. 3. How did we approach or process evolve while working on this launch? Our approach evolved significantly across three key realizations: From "Black-Box AI" to Teacher Agency: Initially, we explored automated AI evaluations, but early feedback made it clear: educators must never feel replaced or sidelined. We shifted control back to the instructor—enabling custom system prompts, three calibrated strictness levels (Flexible, Normal, Strict), and transparent partial-credit logic. From Individual Competition to Collective Defense: We originally built traditional leaderboard contests, but found that authentic problem-solving flourishes under psychological safety and collaboration. This led us to develop cooperative game mechanics like The Great Wall (where correct submissions rebuild a shared class defense) and Group Bucket (where peer teams balance strategy and accountability). From Standalone Tool to Rigorous Standards: As our institutional footprint grew, data privacy and pedagogy took center stage. We aligned the platform directly with the UNESCO AI Competency Frameworks and EU DigComp 2.2, backed by strict row-level security and single-session enforcement, ensuring that innovation never comes at the expense of student safety.

Saval

Gamified PBL & AI Formative Assessment for Educators

Gamified PBL & AI Formative Assessment for Educators