AI Engineering
Learn LLMs, RAG, agents, secure tools, and production AI through practical insurance examples.
Core vocabulary
Use this after studying the full explanations. Spend 10 minutes on vocabulary, 10 on RAG and agents, 10 on security/workflows/evaluation, and 10–30 on the insurance design and scenario answers.
RAG revision
Ingestion: approved source → secure parse/OCR → coherent chunks + metadata → embeddings → complete published index.
Agents and frameworks
LangGraph essentials: nodes return state updates; reducers define merge behavior; edges control next steps; model routes need validation; production needs durable checkpoint storage and authorized threads. An interrupted node can restart, so pre-pause side…
Workflows, memory, and safe operations
- Deterministic: known rules and transitions. Best for permission, arithmetic, approvals, and business commits.
Security and guardrails
Secure action flow: LLM proposal → allowlisted tool → identity/authorization → schema + rules → limits → business service → minimal result.
Evaluation and observability
Evaluate no evidence, stale policies, malformed documents, ambiguous dates, denied access, injected content, restarts, duplicate submissions, and uncertain writes. Track source/model/prompt/graph versions, p95 latency, tokens, cost per successful task,…
Cost, multimodal, and deployment
Cost formula: input tokens × input rate + output tokens × output rate + tools/retrieval/embedding/infrastructure/review, with applicable caching and provider accounting. Prices must be checked for the real provider/model/account.
Two-minute insurance system answer
“Next.js provides the interface and Node authenticates, validates, and applies budgets. A bounded router selects policy RAG, exact claim lookup, guided intake, or staff escalation. RAG returns applicable policy evidence with citations; tools enforce…
Questions to ask before choosing architecture
What successful outcome does the client need, and can ordinary code provide it?
Difficulty is an editorial estimate. Completion means you chose to mark a question as understood.