01
API contracts
Define endpoints, payloads, status states, and error responses before implementation.
/be
Backend AI Engineering
Build reliable server-side AI systems (retrieval, structured output, and eval harnesses) that survive real use.
Overview
What this track proves.
Backend AI
This track is for people who like systems, data, and the parts of AI products users do not always see. You will design API contracts, retrieval-backed answer flows, structured outputs, tool-calling workflows, and evaluation harnesses while owning correctness and failure handling.
Pace
6-10 hrs/week
Prerequisites
Comfort with one programming language helps. Production backend experience is not required.
Capstone
RAG prototype, structured-output pipeline, agent workflow, or reliable backend service.
Review
Async review on contracts, eval evidence, grounding, and failure modes.
Good fit
This is for you if...
You like understanding how data and models move through a product.
You care about grounding, validation, audit trails, and failure handling.
You want to build RAG, agent, or API services that other people or interfaces can depend on.
Outcomes
What you can leave with.
The goal is proof of work, not passive course completion.
Design an AI backend workflow with inputs, outputs, constraints, and failure cases.
Build or specify a retrieval-backed answer flow, structured-output pipeline, or agent loop.
Create a small evaluation set and compare outputs against a rubric.
Write verification notes on quality, limitations, and recommended improvements.
Optional Anthropic courses when partner access allows. Read the FAQs. Your capstone remains the main credential proof.
Curriculum
The work, in order.
02
Task design and prompting
Turn an ambiguous AI idea into a specific job with boundaries, schemas, and success criteria.
03
Retrieval and grounding
Add source context when needed and evaluate whether the answer is actually supported.
04
Evaluation and operations
Build a lightweight rubric, handle auth and retries, and document what you would monitor in production.
Capstone examples
The artifact can take a few shapes.
A RAG prototype with citations and an evaluation table.
A structured-output workflow that extracts, classifies, or drafts useful work with tests.
A simple agent or tool-calling workflow with failure-mode notes.
A deployed API, worker, or integration with audit logging and a verification checklist.
Builder signal
What Builder-level work looks like.
Builder means you have shipped a deployed backend with RAG, structured output, or agent design, evaluation evidence, and a clear explanation of failure modes.
Tool access
What you need to start.
The track is designed around accessible tools and clear alternatives. Use this as a practical setup check before applying.
ToolAccessAlternatives and caveats
Server runtime
Required
Next.js route handlers, Express, FastAPI, or another HTTP backend
Use the stack you can deploy and explain clearly.
Model access
Required
OpenAI, Anthropic, Google AI Studio, local models, or a course-provided tool
Use low-cost or free options where possible; document any paid API use.
Database
Required
Supabase free tier, Neon, SQLite, or a hosted Postgres project
A small persistent dataset is enough for most capstones.
Evaluation workspace
Required
Google Sheets, CSV, notebook, or a small test script
A modest eval set is better than a large demo with no quality check.
Next.js route handlersSupabasePostgresOpenAI or Anthropic modelsEmbeddings or retrieval toolsZodSpreadsheets for evals