Modedu: an EdTech platform rebuilt AI-native
Modedu is an EdTech company. We rebuilt its product as a custom SaaS platform with its own admin panel, ran agentic workflows through the operations behind it, and put AI inside the product itself: academic papers personalised to the student, and a custom whiteboard designed around agents from the outset.
A product Modedu owns, with AI inside it rather than beside it
Modedu needed a custom SaaS platform of its own, with an admin panel for the team that runs it. It needed the operations behind that product to run as agentic workflows, with agents taking the routine work and people taking the exceptions. And it wanted AI in the product in two forms: academic papers personalised to each student, and a whiteboard where the agent is a participant from the first design sketch rather than a feature added to a finished drawing tool.
Four deliverables, one product. On a build like this the platform comes first, because the agentic operations and the AI features both depend on its data.
What we design in by default
- The client owns the IP from day one of any paid stage, code and infrastructure definitions included
- Model-heavy work runs off the request path, behind queues and schedules, so a slow model call never blocks a user
- An agent hands work back to a person wherever an approval step is realistically appropriate
- Deployed in the AWS region you need; for Australian data residency that is the AWS Sydney region
- Documentation and handover are included, with the option to take the software in-house
At a glance
- Client
- Modedu
- Industry
- EdTech
- Scope
- Custom SaaS platform and admin panel, agentic workflows across operations, AI for personalised academic papers, agentic-first custom whiteboard
- Outcome
- 2× revenue within 6 months; $400k → $1M ARR. Two separate client-reported figures
Platform, operations and AI in the product
Three of the four deliverables. The whiteboard has a section of its own below.
Custom SaaS platform and admin panel
The product Modedu's customers use and the admin panel its team runs it from, built as one custom platform that Modedu owns outright. We design the admin panel alongside the customer-facing product rather than after it: the operations team is the first user of every workflow, and an admin panel built late is where manual work hides. This is the work our SaaS development service exists for.
Agentic workflows across operations
The recurring work behind the product runs as agentic workflows. The pattern we build to is plain: an agent takes the task it can complete and verify, and hands the rest back to a person through an explicit status rather than a silent failure. Human oversight is designed in, not added when something goes wrong. See how we build this kind of system under AI agent development.
AI for personalised academic papers
Inside the product, AI generates academic papers personalised to the student. We treat this as AI feature development in the strict sense: the feature is designed into the product rather than bolted on beside it, grounded in the product's own data, with evaluation designed in from the first block. The model behind it is a component we select for the task, not the product.
A whiteboard designed around agents from the outset
An agentic-first whiteboard treats the agent as a participant on the board, not a button beside it. For Modedu that decision was made at the start of the design rather than after a drawing tool was finished. The questions we settle first on a build like this are what the agent may do on the board, how its actions are checked before they land, and where the student stays in control. Only then does the interface get built around the answers.
Agents on a canvas are a harder problem than agents in a chat window. A reply in words can be read and set aside; a reply drawn on a shared board changes what the student is looking at. That is the reason to design around the agent from the outset instead of adding it later, and it is why this piece of the work is custom software rather than a plug-in.
What agentic-first means in practice
- The agent is designed in as a participant; the student keeps control of the board
- The agent's actions on the board are constrained and checked, not free-form
- The model behind the agent is a component chosen for the task, and can change as models change
- Custom software, owned by Modedu from day one of the paid work
How we build an AI-native SaaS platform
The engineering pattern we design platforms like Modedu's around. The specifics of any client's system stay with the client; the approach is ours to share.
Serverless and event-driven on AWS
We design platforms of this kind as serverless, event-driven systems on AWS: managed compute, queues and schedules for anything that runs longer than a web request, and infrastructure described as code so the whole environment can be rebuilt from the repository. It usually lives in the client's own AWS account. Our AWS consulting covers this side of the work.
Models as components, not the product
Model providers are components we select and engineer around. We route model calls through one place, fix the model and token budget per task, and keep the option to change providers as models change. Retrieval, evaluation harnesses and tracing sit around the model so output quality is measured rather than assumed. That is the substance of our production AI engineering.
Agents with a way back to people
We design agents with an explicit hand-off by default: least-privilege access, actions validated before they take effect, and a status a person can see and pick up. An agent that stops and asks is doing its job. This is the agentic workflow automation we run through a client's operations.
2× revenue within 6 months
Modedu reports 2× revenue within 6 months. As a separate figure, it reports annual recurring revenue growing from $400k to $1M. Both are client-reported; we publish them as reported and do not combine them.
What sits behind the numbers is the product itself: a custom SaaS platform and admin panel Modedu owns, operations that run as agentic workflows with people handling the exceptions, AI-generated papers personalised to the student, and a whiteboard built around agents from the outset. Implementation at ComplxAI runs in 12-week blocks with working software every couple of weeks; the full five-stage engagement is set out under how we work.
The services behind this build
AI product development
Products where AI is the point, from first design to production.
AI-native product buildsSaaS development
Multi-role platforms with billing, an admin panel and an operations back end.
Custom SaaS platformsAI agent development
Tool-using agents with guardrails and a hand-off to people.
Custom AI agentsAI feature development
Grounded AI features added to a product you already run.
AI features for an existing productThree more systems with client-reported outcomes
Luxpip
FinTech trading. The same move as Modedu, agents behind the product, applied to a trading platform with web and mobile apps, an admin panel and a B2B API; the client reports manual workload down 75%.
Read the Luxpip case studyDisabilityAssessments
NDIS allied health. Automation rather than agents: onboarding, matching, alerting, invoice processing and reconciliations, with the client reporting 90% of operations automated.
Read the DisabilityAssessments case studyBilly
The closest cousin to Modedu's operations in this set: a custom platform with agentic workflows running through its core operations.
Read the Billy case studyRebuilding a product around AI?
Book an AI strategy call. The first two conversations are free, every stage after that is priced in writing before it begins, and you can stop at any boundary.