Case study · Career guidance platform

GradNextSteps

AI-enabled global education guidance platform.

Industry
Education and career guidance
Current status
MVP — In development
Focus areas
Education, AI Platforms, Product Strategy
External links
Not yet public

01 · Executive summary

In brief

GradNextSteps structures how students, advisers and institutions work through post-graduation decisions on a single shared record. The commercial argument is adviser capacity: preparation work is absorbed by the platform, so adviser time is spent advising rather than reconstructing context.

02 · Business context

The business and its environment

Education guidance is delivered by a small number of experienced advisers against a large, seasonal student intake across multiple countries. Context lived in spreadsheets, email threads and documents held by whoever last touched them.

03 · The challenge

The problem that existed

Students and advisers coordinated high-stakes decisions across spreadsheets, documents and email. No party had a shared view of progress, and advisers spent most of their time reconstructing context rather than advising.

04 · Product vision

Why the product was created

The product exists to make adviser expertise scale without diluting it. The decision was to standardise the process into stages and let AI handle preparation, rather than automate advice — which no institution would accept and no student should receive.

05 · Solution

The platform and the approach

A guided decision workflow that breaks the process into defined stages, holds each student's context in one record, and gives advisers and institutions a current view of where every student stands and what is blocking them.

06 · Architecture overview

Structural decisions

The decisions that were expensive to change later, and were therefore made first.

  • Server-rendered web application for fast first load and full crawlability of public content
  • Stage-based workflow engine with per-student state and adviser assignment
  • Role-separated access: student, adviser and institution administrator
  • Event log capturing stage transitions and adviser actions for accountability
  • Document handling isolated from application data, with signed, expiring access
Architecture diagram reserved. It shows the stage workflow engine, the single student record, the three access roles, and the document store held apart from application data behind signed, expiring access.

07 · Key features

What the platform actually does

Listed as capabilities that change how work happens, not as a feature inventory.

Stage-based decision workflow

The process is broken into agreed stages, so progress is visible to the student, the adviser and the institution at the same time.

Single student record

Goals, documents, prior conversations and decisions sit in one place, which removes the reconstruction work before every session.

Adviser briefing before each session

A short summary of the student's stated goals and prior documents is prepared ahead of the session, with every statement traceable to its source.

Role-separated access

Student, adviser and institution administrator see different views of the same record, enforced below the application layer.

Institutional progress view

Administrators see where cohorts stand and what is blocking them, without asking advisers for status reports.

Accountability log

Stage transitions and adviser actions are recorded, so an institution can evidence the guidance a student received.

08 · Technology stack

What it runs on

Conventional choices, selected so a team other than the one that built it can run and extend it.

Application
  • TypeScript
  • React
  • Server-side rendering
Data
  • PostgreSQL
  • Row-level access policies
Platform
  • Managed Postgres and auth
  • Edge delivery
AI
  • Hosted language models behind an internal boundary

09 · AI usage

Where AI adds value, and where it does not

Language models summarise a student's stated goals and prior documents into a briefing an adviser reads before a session. Output is advisory only, always attributed to its source material, and never presented to students as a recommendation without adviser review.

10 · Security and compliance

What was decided before the build

Included because these decisions are difficult and expensive to retrofit.

  • Role-level access enforced by database policy, not by interface state
  • Student documents held apart from application data, reachable only through signed, expiring links
  • Event log covering stage transitions and adviser actions, for institutional accountability
  • Cross-border data handling considered at design time, given the international intake

11 · Current status

Where it stands today

MVP

In development. The measures being tracked are adviser preparation time per student, stage completion rate, and the proportion of sessions that begin with context already available.

12 · Lessons learned

What the work taught

Strategic rather than technical: the judgements that carry into the next engagement.

  • Advisers adopt a system when it removes preparation work, not when it adds reporting.
  • Stage definitions must be agreed with advisers before any interface work — naming drives behaviour.
  • AI summaries are only trusted when the source of each statement can be opened directly.

Interface

Product views

Reserved for published screenshots. Alt text and captions are in place so the section is meaningful once images are added.

Adviser view: students grouped by stage with blockers surfaced first.
Guided stage with the student's context and prior documents alongside.

15 · Next evolution

What comes next, and why

Sequenced against business value rather than technical interest.

  1. Institution-configurable stage definitions, so a university can model its own guidance process
  2. Outcome tracking after placement, connecting guidance quality to real destinations
  3. Adviser capacity planning informed by stage throughput rather than headcount assumptions

The best products begin with the right questions.

Send the business context and the constraint you are working against. I will reply with a direct read on the approach I would take, and whether I am the right person for it.