G01 Stage C · Applied GenAI 44 h (16 T / 28 P)
Large Language Models & Prompt Engineering
Builds reliable, schema-valid applications on top of language models.
Prerequisite
C04
Modules
5
NOS
—
Statutory mapping
—
Tech stack
OpenAI APIAnthropic APIPydanticLiteLLM
Modules
5 modules · 44 hours
M1 · LLM landscape
8 h TOPICS
- Pretraining
- Instruction tuning
- Alignment (RLHF/RLAIF)
- Model families
- Proprietary versus open-weights
- Context windows
- Capability and limitation profiles
- Model selection criteria
Lab: Benchmark three models on one task for quality, latency and cost
Course material for this module is in production.
M2 · Inference mechanics
8 h TOPICS
- Tokenisation and token accounting
- Temperature, top-p, top-k
- Stop sequences
- Determinism
- Max tokens
- Streaming
- Context-window management and truncation strategy
Lab: Token-budget calculator and sampling sensitivity study
Course material for this module is in production.
M3 · Prompt engineering
10 h TOPICS
- Zero-shot and few-shot
- Chain-of-thought
- Role and system prompts
- Delimiters
- Output contracts
- Self-consistency
- Decomposition
- Prompt templating and versioning
- Anti-patterns
Lab: Build a versioned prompt library with regression cases
Course material for this module is in production.
M4 · Structured outputs and integration
10 h TOPICS
- JSON mode
- Schema enforcement with Pydantic
- Function and tool calling
- Validation and repair loops
- Error taxonomy
- Retries and idempotency
- Provider SDKs and gateways
Lab: Service returning strictly schema-valid structured output with repair
Course material for this module is in production.
M5 · Hallucination
8 h TOPICS
- Causes
- Detection
- Grounding strategies
- Refusal design
- Confidence signalling
- Citation requirements
Lab: Measure and reduce unsupported-claim rate on a fixed task set
Course material for this module is in production.
Real-world work scenario
A customer-facing summariser occasionally invents policy numbers. Build a reproducible failure set, add grounding and schema validation, define refusal behaviour, and demonstrate a measured reduction in fabricated fields.
Assessment
Prompt library with regression cases · structured-output service