StudyRadar

The problem.

StudyRadar prepares candidates for medical examinations. Production failures spanned four independent surfaces — geographic access, Firebase OAuth, Stripe payments and OpenAI API connectivity — each interrupting a different step of a paying candidate's journey. The platform also needed a way for candidates to rehearse spoken OSCE-style examinations rather than only reading material.

My part in it.

Backend and DevOps engineering: diagnosing and resolving the production failures, then building and deploying the real-time voice practice feature and the session and referral systems supporting it.

The shape of the challenge.

Real-time spoken practice runs against OpenAI's live API and was deployed on dedicated VPS infrastructure provisioned for it. Exam session events had to be authenticated so that downstream systems could trust them.

What went into it.

1. 01

   Resolved production failures across geographic access, Firebase OAuth, Stripe payments and OpenAI API connectivity for a medical exam preparation platform.

2. 02

   Built and deployed real-time AI voice practice for OSCE exam simulations, integrating OpenAI's live API and provisioning dedicated VPS infrastructure.

3. 03

   Implemented authenticated exam-session webhooks and a referral program with configurable discounts and credits.

What it turned on.

• Repair the candidate journey end to end rather than one outage at a time.

  Geographic access, Firebase OAuth, Stripe payments and OpenAI connectivity each interrupted a different step of the same paying candidate's journey.

• Provision dedicated VPS infrastructure for the voice sessions.

  Real-time spoken practice runs against OpenAI's live API and needs a session that stays open.

• Authenticate exam-session webhooks.

  Downstream systems had to be able to trust session-completion events.