The agent grades itself.

PASS·Mandatory

Tell me about yourself — where are you based, and what brought you to AI engineering?

PASS·Mandatory

Walk me through your education — you double majored in CS and Data Science at UW-Madison?

PASS·Nice-to-have

You're open to relocation — what would make you leave NYC?

PASS·Out-of-scope

What's your target salary range for your next role?

PASS·Mandatory

You've been at Enidus for about 9 months — what's the scope of your role as a Software Developer?

PASS·Nice-to-have

What does a typical week look like for you at Enidus?

PASS·Nice-to-have

How much autonomy do you have to make architectural decisions on these three systems?

PASS·Mandatory

Tell me about the architecture of the T-Mobile AI copilot — how are the three layers (intent, tools, execution) separated?

PASS·Mandatory

You mention 43 Pydantic-typed tools — what does that mean, and why did you choose that constraint?

PASS·Mandatory

Your eval suite caught the agent hallucinating tool arguments — can you walk me through what you found and how you fixed it?

PASS·Mandatory

Tell me about the eval suite — 52 cases covering what exactly, and how did it shape the architecture?

PASS·Mandatory

Why did you go with 53 hand-designed intents instead of free-form NLU?

PASS·Mandatory

You mention 8-role RBAC — what are those roles, and how does the session token determine what tools the model can see?

PASS·Mandatory

You use Qdrant for the knowledge layer — why per-tenant collections instead of a shared collection with a tenant filter?

PASS·Mandatory

Walk me through the three layers of multi-tenant SQL isolation you've implemented.

PASS·Mandatory

Why did you choose stage-and-confirm over letting the model write directly? What was the first version like?

PASS·Mandatory

You're in pilot with 15 reseller tenants representing 25+ enterprise customers — what have you learned from pilot users?

PASS·Mandatory

Tell me about the Custom Reports & Dashboards platform you built end-to-end alone — what does it actually contain?

PASS·Mandatory

What was the business impact of this product on analytics turnaround and engineering time?

PASS·Mandatory

Walk me through all four security layers of the reports platform.

PASS·Mandatory

Why did you use AES-256-CBC encryption for stored Excel passwords? What's the threat model?

PASS·Mandatory

You mention tenant clause bound at runtime, not save time — why does that distinction matter?

PASS·Mandatory

Cron jobs run through the same code path as interactive runs — why is this a security decision?

PASS·Nice-to-have

You say 'stored procedures as the contract — no raw SQL from controllers' — what's the advantage over parameterized queries?

PASS·Mandatory

You mention establishing the Enterprise Portal plugin pattern — what is that?

PASS·Mandatory

You mention a third production system — Carrier API Gateway / BFF — that's not on your portfolio. Why?

PASS·Mandatory

What's the tech stack of the Carrier API Gateway?

PASS·Nice-to-have

Tell me about the OAuth and PoP token generation in your BFF — what security properties are you achieving?

PASS·Nice-to-have

How do you handle carrier API failures in the BFF? What's your retry and fallback strategy?

PASS·Mandatory

Describe the meta-loop of ClaudeJob — how does it power your own applications, and what's the validation stack?

PASS·Mandatory

How does source-fact validation work? What happens if the LLM invents a statistic?

PASS·Mandatory

You mention 'deterministic adjacency-skill injection' — what does that mean?

PASS·Mandatory

Why did you choose JSON as the LLM contract instead of plain-text resumes?

PASS·Nice-to-have

You switched from Python subprocess PDF rendering to pdfkit — what problem did that solve?

PASS·Nice-to-have

You have 47 passing unit tests for ClaudeJob — what are they testing?

PASS·Mandatory

Walk me through your Denari RAG capstone — corpus size, architecture, retrieval approach, and results.

PASS·Mandatory

Why hybrid retrieval instead of pure semantic search?

PASS·Nice-to-have

Why TimescaleDB hypertables instead of a generic vector DB?

PASS·Nice-to-have

You mention 'Agile/Scrum delivery on a 4-month clock' — how did you scope 25+ features?

PASS·Nice-to-have

What's the single biggest thing this project taught you about RAG systems?

PASS·Nice-to-have

Why do you include a weather pipeline project from coursework on your portfolio?

PASS·Nice-to-have

You chose Cassandra over PostgreSQL/TimescaleDB for the weather pipeline — why?

PASS·Mandatory

Partition key design was the most important decision in the weather system — tell me about it.

PASS·Nice-to-have

You chose gRPC over REST for ingest — what did that buy you?

PASS·Nice-to-have

Tell me about your fault-tolerance testing — how did you verify the replication model actually held?

PASS·Nice-to-have

How does this weather pipeline project directly inform your production thinking now?

PASS·Mandatory

Tell me about your Orahi internship — the bus route optimization and what you learned.

PASS·Nice-to-have

For a medical imaging task like pneumonia detection, precision vs recall is critical — how did you think about the tradeoff?

PASS·Nice-to-have

Why include a high-school internship on your resume if most of your real engineering experience is post-2023?

PASS·Nice-to-have

You have SnowPro Associate & Core certification from 2024 — what prompted that?

PASS·Nice-to-have

Stored procedures add friction to the application layer — why force CRUD through them instead of ORM + parameterized queries?

PASS·Nice-to-have

Is there anything in your projects you'd describe as overengineered in retrospect?

PASS·Nice-to-have

If you were rebuilding the T-Mobile copilot from scratch today, what would you do differently?

PASS·Nice-to-have

Why Claude for the ClaudeJob pipeline instead of GPT-4 or Llama, and are you happy with that choice?

PASS·Mandatory

Walk me through your eval philosophy — when do you write eval cases, and how do you measure what matters?

PASS·Mandatory

When the copilot was hallucinating tool arguments, did you consider a prompt fix first, or go straight to structural?

PASS·Nice-to-have

When would you choose NOT to use an LLM for a task that involves language?

PASS·Nice-to-have

What's your opinion on RAG vs fine-tuning for grounding LLMs to domain-specific knowledge?

PASS·Out-of-scope

Have you considered or used agentic frameworks like LangChain, AutoGPT, or ReAct, and what's your take on them?

PASS·Nice-to-have

Tell me about a time you disagreed with a senior engineer or manager on an architectural decision.

PASS·Nice-to-have

What's the hardest bug you've had to track down, and how did you solve it?

PASS·Nice-to-have

How do you handle ambiguity when requirements aren't clear?

PASS·Nice-to-have

Tell me about a time you had to push back on scope or timeline because the quality bar wasn't met.

PASS·Nice-to-have

Tell me about something technical you've learned in the past six months that changed how you think about engineering.

PASS·Nice-to-have

You're applying to both AI Engineer roles at 5-30 person startups AND traditional SWE roles — what's the difference you're looking for?

PASS·Nice-to-have

If you joined our 8-person AI-first startup next week, what would you want to ship in your first 90 days?

PASS·Out-of-scope

Where do you see yourself in five years?

PASS·Nice-to-have

You have a lot of small, well-executed projects but nothing in production at scale for years — is that a concern?

PASS·Nice-to-have

You didn't go deep on graduate-level ML research — does that limit your ability to reason about advanced ML topics?

PASS·Out-of-scope

Why leave Enidus, which is shipping real products, to join a startup with more risk and less stability?

PASS·Nice-to-have

What technical skill do you not have yet that would make you more effective at an AI-first startup?

PASS·Out-of-scope

What do you know about our company specifically, and why are you applying?

PASS·Out-of-scope

If we had $5 million in funding and you picked the problem space, what would you build?