Shelf · AI
AI (capped)
Capped on purpose. LLM APIs, RAG, evals, agents and FastAPI serving: enough to build and to judge, no hype.
1
LLM mental model
PlannedDifficulty
Priority
- Tokens and the finite context budgetEasyP3Talk about it
- Embeddings are similarity representationsEasyP3Talk about it
- Temperature and top-p have model limitsMediumP4Talk about it
- Why plausible output can be falseEasyP2Talk about it
- Model families and workload trade-offsMediumP4Talk about it
- When not to use an LLMEasyP2Talk about it
2
Prompt engineering as engineering
PlannedDifficulty
Priority
- Prompt engineering with versions and evalsMediumP3Talk about it
- Roles, structure and delimitersEasyP3Know cold
- Few-shot examples with negative casesEasyP3Talk about it
- Chain-of-thought caution and checkable outputsMediumP5Know it exists
- JSON requests versus schema contractsMediumP3Know cold
- Injection basics and adversarial prompt evalsMediumP3Talk about it
3
LLM APIs in Python
PlannedDifficulty
Priority
- LLM API integration for Claude and OpenAIMediumP3Know cold
- Streaming as an owned request lifecycleMediumP3Know cold
- Tool and function calls are proposalsMediumP3Know cold
- Structured outputs at the Python boundaryMediumP3Know cold
- Retry, timeout and rate-limit budgetsHardP3Know cold
- Token accounting, cost control and prompt cachingMediumP4Talk about it
4
RAG that actually works
PlannedDifficulty
Priority
- Retrieval-augmented generation and chunk ownershipMediumP3Talk about it
- Embedding and vector-store selectionMediumP3Talk about it
- Hybrid search for names and meaningMediumP3Talk about it
- Rerank candidates before spending contextMediumP4Talk about it
- Citations that actually support a claimMediumP3Talk about it
- Diagnose retrieval separately from generationHardP4Talk about it
5
Agents & tools
PlannedDifficulty
Priority
- Agent tool loops with bounded recoveryHardP3Talk about it
- Planning and durable task stateMediumP3Talk about it
- Tool contracts and executable guardrailsMediumP3Know cold
- MCP protocol, capabilities and trustMediumP3Talk about it
- Human approval tied to the exact actionMediumP3Talk about it
- Prove when multi-agent is justifiedHardP4Know it exists
6
Evaluation & reliability
PlannedDifficulty
Priority
- LLM evaluation and golden datasetsMediumP4Talk about it
- LLM-as-judge without treating it as truthHardP4Talk about it
- Faithfulness, groundedness and correctnessMediumP4Talk about it
- Prompt and model regression gatesMediumP4Talk about it
- Offline versus online evaluationMediumP4Talk about it
- Trace observable LLM behavior safelyMediumP3Talk about it
7
Productionising an AI feature
PlannedDifficulty
Priority
- Architecture, async and queued AI workHardP3Talk about it
- Stream a UI without exposing partial actionsMediumP3Talk about it
- Cache answers without crossing permissionsHardP3Talk about it
- Prompt-injection defense and tool permissionsHardP3Know cold
- PII, privacy and data governanceHardP2Talk about it
- Cost ceilings, fallbacks and reversible rolloutHardP3Talk about it
8
AI-assisted engineering, honestly
PlannedDifficulty
Priority
- AI-assisted engineering without surrendering judgmentMediumP2Talk about it
- Spec-first workflow and acceptance evidenceEasyP2Know cold
- Review generated code as accountable engineeringMediumP2Know cold
- Where AI made work measurably slowerMediumP5Know it exists
- Preserve learning and measure real outcomesEasyP5Know it exists
- What the 2026 market pays for: evidence limitsEasyP5Know it exists
No cards match these filters.