●Blog · 15 posts

AI engineering, explained simply.

Notes from preparing for an AI Technical Manager interview: every question I expected, answered the way I'd say it out loud, with the architecture behind it.

03Interview questions

Twelve questions, answered plainly.

MCP, agents, RAG, evals, vibe coding, cost, multi-tenancy and leadership. Each post has a diagram and the short version to say out loud.

2 min read
Q1MCP

Q1. How do you design an MCP server: tools, resources or prompts?

How I design Model Context Protocol (MCP) servers for enterprise systems: deciding between tools, resources and prompts by who is in control, using Streamable HTTP, and filtering tools so the model never sees 100 at once.

2 min read
Q2MCP

Q2. How do you secure MCP servers?

Securing Model Context Protocol servers in an enterprise: OAuth 2.1 with audience checks, no token passthrough, scoped downstream credentials, treating tool output as untrusted, and human approval for risky actions.

2 min read
Q3Agents & RAG

Q3. Simple workflow or autonomous multi-agent system?

When to use a single LLM call, a deterministic workflow, an agent graph or a multi-agent system, and the guardrails (step limits, token budgets, checkpoints, human handoff) that keep agents from running away.

2 min read
Q4Agents & RAG

Q4. Your RAG system is slow and makes things up. How do you fix it?

How I fix a RAG pipeline that is slow and hallucinates: hybrid vector and keyword search, tenant pre-filtering, reranking, Redis caching, citations and an abstain threshold.

2 min read
Q5Agents & RAG

Q5. How do you test and monitor AI in production?

A practical setup for AI evaluation: a golden test set run on every change, shadow and canary rollouts, and production signals like time to first token, cost per session and how often users edit AI output.

2 min read
Q6Engineering

Q6. How do you let teams vibe code without wrecking code quality?

How I roll out AI coding assistants to an engineering team: shared repo rules, tests first, security scanning in CI with tools like Vibefort, and the rule that the engineer owns every line they merge.

2 min read
Q7Architecture

Q7. How do you keep LLM cost and latency low as traffic grows?

Three levers for LLM unit economics in a SaaS product: route easy tasks to small models, cache with prompt caching and Redis, and stream responses from FastAPI while pushing post-processing to a queue.

2 min read
Q8Architecture

Q8. How do you make sure customer A never sees customer B's data?

Tenant isolation for AI backends in Python and TypeScript: tenant_id from verified auth, Postgres row-level security, pre-filtered vector search, namespaced Redis keys, and never letting the LLM choose the tenant.

2 min read
Q9Leadership

Q9. How do you turn full-stack developers into AI engineers?

How I grow backend and frontend developers into AI engineers: a ready starter template, a step-by-step path from structured output to RAG, agents and MCP, and a weekly review of real failures.

2 min read
Q10Leadership

Q10. Product wants autonomous agents, Sales wants a demo next week. What do you do?

How I handle competing pressure from Product and Sales on an AI agent feature: a guided flow with human approval in two weeks, evals from real usage next, and autonomy only where accuracy proves it.

2 min read
Q11Leadership

Q11. Tell me about an AI system you built and the value it created

Two AI systems I architected end to end: ResolveOps, an AI billing-support agent with cited investigations and human-approved refunds, and the Actionable HQ commitment engine with WhatsApp follow-ups for 10,000+ users.

2 min read
Q12Leadership

Q12. What do you look for when hiring AI engineers?

What I test for when hiring AI engineers: software fundamentals, structured debugging of real AI failures, and product sense about cost, speed and what happens when the AI is wrong.