Six schools. Choose your specialisation.

“Data is the foundation. Everything else is built on top of what you build here.”

“Tell me WHY before you show me HOW.”

“Before you write the query, draw the tables. What connects them?”

“What happens when this task fails at 3am? Show me your retry logic.”

“Model for the questions people will ask, not the data you have.”

“A model that works but can’t be explained is a liability, not an asset.”

“Before you tune hyperparameters, tell me why you chose this architecture.”

“Try a simple baseline first. You’d be surprised how often it wins.”

“Think of attention as the model asking: which parts of the input matter most for this output?”

“Your model is accurate. Now make it run in 50ms on a device with 2GB RAM.”

“AI engineering is not about the model. It’s about the system around the model.”

“But does it scale?”

“If your retrieval is wrong, your generation is confidently wrong. Fix retrieval first.”

“Show me your training data before you show me your hyperparameters.”

“What’s the cost per request? Your architecture is only viable if the unit economics work.”

“You’re not building a tool. You’re building something that makes decisions. Treat that seriously.”

“What happens when agent A and agent B disagree? Design for conflict, not just cooperation.”

“Build a tiny agent first. Make it work. Then make it smart.”

“An agent without memory is just a function call. Memory is what makes it an agent.”

“If it’s not in production with monitoring, it doesn’t exist.”

“Your model is only as reliable as your deployment pipeline. Show me the pipeline.”

“What’s your p99 latency? What’s your cost per prediction? Those two numbers define your architecture.”

“If your alert fires, you’re already late. Design monitoring that predicts the problem.”

“The question is never just "can we build it?" It’s "should we, and how do we build it responsibly?"”

“Compliance is not paperwork. It’s architecture. Build it into the system from day one.”

“If a patient asks why the model flagged them, what do you say? That’s explainability.”

“I found three ways to make your model produce harmful output. Now let’s fix all three.”

“Would you deploy this to production with your name on it?”

“This isn’t a coding exercise. The retrieval team is blocked until your pipeline ships.”

“Your chunking strategy tells me everything about how you think about this problem.”

“An agent that can’t explain its own decision is an agent you can’t trust.”

“What’s the smallest thing you can ship today that moves us forward?”

“Show me how this handles a million records. Then we’ll talk about your algorithm.”

“If you can’t monitor it, you can’t ship it.”

“I evaluate idempotency before business logic.”

“A notebook is not a deployment. Show me the Docker file, the CI pipeline, and the rollback strategy.”

“If your teammate can’t understand this code in 6 months, rewrite it.”

“Fast enough is never fast enough. Find the bottleneck and eliminate it.”

“You have 2GB of RAM and 100ms latency budget. Make it work within those constraints.”

“Explain the receptive field of your architecture. If you can’t, you don’t understand your model.”

“The best architecture is the simplest one that meets your requirements. Start there.”

“What’s the smallest thing you can ship today?”

“Behind every data point is a patient. Build systems worthy of that responsibility.”

“Every line of code in healthcare is auditable. Write it like a regulator is reading it.”

“In healthcare, a logging mistake isn’t a bug — it’s a lawsuit.”

“Your model denied a patient coverage. Can you explain why to their doctor? That’s the standard.”

“We ship slower than other companies. But what we ship never harms a patient.”

“Your profile says more than your resume ever could.”