EDTECH & KNOWLEDGE-SHARING PLATFORM
How full-stack engineering turned a broad vision into a scalable education and workplace AI platform
Introduction
Superintelligent began with a bold idea: to make AI education more practical, more social, and more accessible. What started as a learning platform for individuals quickly turned into a system used by companies to collect and share AI use cases across teams.
As a founding engineer, I helped define the technical direction of the platform—architecture, infrastructure, backend, search, streaming, and many user-facing features.
This case study tells the story of how we built a scalable edtech application, navigated a major product pivot, and created an architecture that supported fast iteration, a social content feed, video streaming, and multi-organization permissions.
The Challenge
The earliest version of Superintelligent set out to combine three demanding worlds: video-based learning, social content, and AI assistance. The goal was to create a platform where admins could upload lessons, design challenges, and build learning paths, while students engaged through a real-time feed, project sharing, comments, and an integrated AI coach. Even in its initial scope, the product required a system capable of handling heavy media, continuous interaction, and low-latency responses without compromising on smoothness.
Very quickly, though, we realized the ambition went deeper. As the platform grew, companies began asking whether they could use Superintelligent internally—not just to teach AI skills, but to document and share real AI use cases across departments. That shift introduced a completely different set of constraints. Instead of a public learning environment, organizations needed private workspaces, strict access control, team-specific feeds, and structured repositories where sensitive information could be shared safely.
This pivot transformed the project's technical landscape. We now had to support a multi-tenant architecture, permission-based content visibility, and folder-like structures that aligned with real corporate hierarchies. At the same time, the platform still had to deliver lightning-fast video playback, social interactions, and an interface that felt as immediate and fluid as a consumer application.
Underneath all of this was the constraint that shaped every decision: the platform needed to scale rapidly without ballooning costs or requiring a large DevOps team. The challenge was not simply to build a feature-rich product, but to engineer an architecture robust enough for enterprises and efficient enough for a startup.
Approach
Laying the groundwork for a scalable, cost-efficient platform
To support both the ambition of the consumer-facing product and the emerging enterprise use case, the first priority was to design an architecture that could grow quickly without demanding a large operational footprint. We chose a backend built on Express and Node.js, paired with PostgreSQL on AWS RDS for reliable relational storage. Redis on ElastiCache handled the high-frequency operations needed for feeds, notifications, and caching.
For deployment, we relied on AWS ECS with Fargate. This allowed us to scale automatically during peak usage—such as company-wide onboarding events or heavy video consumption—while avoiding the complexity of managing EC2 instances. It also ensured that we could deploy new versions continuously with zero downtime, a critical requirement for a platform that blended education and social interaction.
Video storage and delivery were handled through Bunny.net, which offered a globally optimized CDN and a cost-efficient streaming pipeline. This decision drastically reduced load times and made video playback smooth even for content-heavy courses that students were likely to consume back-to-back.
With Clerk managing authentication and user accounts, Stripe supporting subscription flows, and Sentry and CloudWatch providing observability, the platform had an operational backbone that balanced flexibility with predictability.
Building a content engine that felt fast and alive
Once the foundations were in place, the next step was to build the core of Superintelligent's learning and social experience. This meant creating a content engine capable of supporting diverse media types—video lessons, polls, articles, challenges, and student projects—while making everything feel instantaneous.
We implemented a pipeline for lesson uploads and video streaming that maintained high quality without introducing unnecessary processing delays. Reliable view tracking ensured accurate analytics for creators and admins.
On top of that, we designed a chronological feed that surfaced the right mix of updates: new lessons, user posts, project showcases, and discussions. Social interactions such as comments, replies, mentions, and likes were engineered to be low-latency, giving the feed the "alive" feeling of a modern social app rather than a traditional LMS.
Admins received a suite of tools for managing courses, designing structured learning paths, organizing content, and tracking user progress—all built with attention to clarity and speed so teams could publish updates as quickly as they produced them.
The UI focused heavily on smooth interactions: hover-to-preview videos, near-instant page transitions, and a layout optimized for both browsing and deep learning. Every detail reinforced the idea that learning shouldn't feel like navigating a corporate portal—it should feel engaging and intuitive.
Search and discovery powered by relevance
As the content library grew, we needed a fast and intelligent way for users to find what mattered to them. We integrated Algolia to deliver sub-100ms searches across lessons, challenges, articles, and user posts. Instead of relying on generic text matching, we built custom ranking strategies that considered content type, user behavior, and learning progress.
This allowed us to provide meaningful recommendations, including contextually relevant "next lessons," without implementing a full recommendation engine. The result was a discovery experience that felt personal and adaptive, even as the platform evolved.
Evolving into a secure, multi-tenant enterprise platform
When organizations began adopting Superintelligent internally, the product's needs shifted dramatically. We redesigned the underlying data models and API structure to support multiple organizations, each with its own users, teams, and content repositories.
This meant creating a system where:
- each organization's data was fully isolated,
- teams could share knowledge while maintaining the right level of visibility,
- content could be organized into folders or collections that mirrored real corporate structures,
- and feeds dynamically adapted to show only what a user was allowed to see.
We implemented a flexible permission model capable of handling everything from private use cases to company-wide resources. The architecture had to be secure, intuitive for admins, and invisible to end-users—all while preserving the platform's signature sense of speed.
This transformation was not a simple add-on; it required rethinking how content flowed through the system and designing an access model that aligned with how teams collaborate in real organizations. The result was a platform that could serve both individual learners and enterprise clients without compromising either experience.
Results
During my time at Superintelligent, we took the platform from an early prototype to a production system used by thousands of users and adopted by companies with hundreds of seats. Teams used it as their internal hub to share AI use cases, document experiments, and roll out AI knowledge across departments.
From an engineering standpoint, the architecture proved itself under real-world load. The combination of ECS/Fargate, RDS, Redis, and Bunny.net handled traffic spikes—such as company-wide onboardings and heavy content launches—without downtime or manual intervention. Video lessons streamed smoothly, feeds stayed responsive, and we were able to deploy new features continuously while users were active.
The multi-tenant and permissions model also validated the original design. Organizations could isolate their data, structure content by teams and folders, and expose the right knowledge to the right people without leaking sensitive information. For admins, it felt powerful and controllable; for end users, it felt like a fast, consumer-grade app rather than a typical enterprise tool.
By the time my work on Superintelligent concluded, we had a proven architecture, a battle-tested product, and a clear demonstration that the platform could support real customers, real workloads, and real usage at scale. As a founding engineer, I was responsible for a large part of that outcome—from the core architecture and infrastructure to the backend, search, streaming, and social features that powered the product in production.
Tech Stack
Want to build something similar?
Let's discuss how we can help bring your edtech platform to life.