Product descriptions:
GoGuardian powers safe, focused learning for half of U.S. K–12 students. Their end-to-end platform streamlines web filtering, classroom management and harm prevention — with AI built for high-stakes moments when every second counts. Despite helping prevent an estimated 18,623 students from harm since March 2020,* GoGuardian was up against many challenges with billions of daily inferences across fragmented infrastructure. They needed a more scalable, cost-effective and integrated platform to manage these high-throughput AI workloads, simplify complex infrastructure and ensure PII-compliant data handling. GoGuardian chose Databricks for their ability to unify data, AI development and deployment — a foundation that helped the EdTech brand achieve a 62% reduction in inappropriate device usage among students.
Facing infrastructure barriers that impacted student safety
GoGuardian was founded on the belief that when thoughtfully applied, technology should enhance student safety, personalize learning and lighten the load for educators. Today, GoGuardian’s AI powers everything from suicide prevention alerts and off-task mitigation to personalized lesson creation and data-informed teaching strategies — helping reduce educator burnout, reach every student and intervene earlier. To do this, GoGuardian uses machine learning (ML) models, including their proprietary model Delphi, to categorize websites and filter content. The software blocks inappropriate or distracting sites (like gaming, music or sports), and detects proxy websites — known as “Smart Alerts” — that students may use to bypass filters.
The company also prioritizes student privacy. They train their models in a PII-free analytics environment that masks or hashes personal information. At the same time, they are experimenting with generative AI (GenAI) tools to identify threats faster and more accurately. Safeguarding students at scale, however, is no easy task. Lives, trust and regulatory compliance are all on the line. GoGuardian must manage millions of ML inferences daily while meeting strict privacy laws and adopting responsible AI practices. On a typical day, the system handles 4 to 6 billion inferences, placing a heavy strain on traditional infrastructure.
Compounding the challenge was the complexity of managing disconnected services across data management, ML development, model serving and monitoring. This fragmentation led to operational and cost inefficiencies, making it difficult to scale AI/ML workloads within their existing environment. “We also wanted to ensure strict compliance with the Children’s Online Privacy Protection Act (COPPA) and the Family Educational Rights and Privacy Act (FERPA),” Manoj Rawat, Director of Data and AI at GoGuardian, said. “That meant building a secure, privacy-first data environment that could redact sensitive information without compromising model training.”
The company’s commitment to responsible AI introduced further complexity. Their systems needed to be explainable, equitable, reliable and aligned with frameworks like the NIST AI Risk Management Framework. Manual data labeling, especially for sensitive use cases like self-harm detection, became a bottleneck, slowing development and underscoring the need for AI-driven solutions to maintain both speed and accuracy. To meet these challenges, GoGuardian turned to Databricks. The platform offered a unified, scalable foundation to modernize their infrastructure, reinforce privacy protections and accelerate AI innovation.
Building AI systems that support distraction-free learning
After migrating to Databricks, GoGuardian quickly saw the benefits of a unified platform that met their real-time performance needs and eliminated the sprawl of siloed AWS services that had previously slowed development and increased costs. They began with Delta Lake to bring structure, reliability and performance to their AWS S3 environment. This created a scalable foundation for the large-scale machine learning and real-time analytics essential to ensuring student safety and learning outcomes.
To take full advantage of this structured data layer, GoGuardian implemented Databricks Lakeflow, building a real-time ingestion pipeline that automated data detection, transformation and validation. This ended manual workflows and ensured that datasets powering their AI models were always fresh and accurate. An integration with dbt enabled key SQL-based transformations within this pipeline, such as hashing logic to remove sensitive student data from downstream workflows.
The team next tackled the complexity of scaling model deployment. By adopting Databricks’ serverless infrastructure, they eliminated the need to manually manage cloud clusters. “For us, the best part about our move to Databricks was no longer having to manage or maintain cloud clusters manually,” Manoj said. “We could rely on Databricks to automatically spin up and scale compute resources based on workload demands.” This shift reduced operational overhead and costs — replacing always-on environments with usage-based compute that scaled as needed.
As GoGuardian’s data maturity advanced, they looked for additional ways to democratize access to insights and enforce centralized control. Databricks SQL enabled teams across the organization to run complex queries without managing back-end infrastructure. Meanwhile, Unity Catalog provided a unified layer of governance — streamlining access controls and enforcing data security policies. These capabilities supported a core tenet of their mission: responsible and ethical AI. With a strong foundation in place for data management and governance, GoGuardian focused on scaling AI development. Using MLflow, the team gained end-to-end control over the machine learning lifecycle — from experiment tracking and model versioning to team-wide reproducibility. When models were ready for deployment, Mosaic AI Model Serving allowed them to scale effortlessly to meet the real-time demands of student monitoring and content filtering, without the infrastructure burdens typically associated with production ML stacks.
One of the company’s most important models, Delphi, powers website classification to help enforce school-specific policies. Throughout their journey, student privacy remained a top priority. With the Databricks Data Intelligence Platform, GoGuardian maintained a PII-free environment for data preparation and transformation, ensuring that only de-identified data reached the training stage. This reinforced their commitment to responsible AI innovation and compliance with strict privacy regulations.
Achieving cost savings, efficiency and student well-being
With their data and AI systems finally working in sync, GoGuardian enjoyed greater agility with their operations, improving their ability to support educator and student-forward initiatives by helping to foster a distraction-free learning environment.
“Our operational costs for machine learning services dropped by up to 50% across key use cases, with some models, like our Delphi website classification model, achieving up to 90% savings compared to previous AWS deployments. This allowed us to reallocate time and effort into building a new GenAI labeling tool,” Manoj concluded.
At the same time, schools using GoGuardian’s filtering solutions reported a 62% drop in inappropriate device use — reducing access to content that violated school policies, disrupted learning or posed risks to student well-being. These improvements have strengthened GoGuardian’s ability to keep students engaged and safe online, already helping prevent an estimated 18,623 students from physical harm since March 2020.* Now, with a secure, PII-free analytics environment embedded into their workflows, GoGuardian is advancing efforts to proactively detect high-risk content. Using AI to prioritize records most likely to indicate threats or violence, they’ve cut the volume requiring human review by over 95% — from 1 million records to just 35,000–45,000 — dramatically accelerating safety model development.
Looking ahead, GoGuardian plans to extend their real-time safeguards with new innovations designed to further protect and support students in digital environments. These include video confidence filters to assess and approve educational content, chatbot detection to flag risky or inappropriate online conversations and smart alerts that predict and block access to gaming sites during school hours. Together, these advancements reinforce GoGuardian’s mission to create safer, more focused digital learning environments where students can thrive without distraction or harm.
*From March 2020 to July 2024, Beacon alerts notified schools and districts of 18,623 students at risk of harm.