INTERMEDIATE · 8 MODULES · 64 LESSONS
Context Engineering
Decide with discipline what goes into a model's context window: budget tokens, order and compress information, and measure the effect of every change.
- For you if
- You build LLM apps on top of documents and want to move from good prompts to budgeting context.
- By the end
- You'll know how to version a system prompt, select content under a token limit, compact long histories, and mitigate context rot.
- You need
- Prompting fundamentals and basic Python installed. No API key or internet connection needed.

Curriculum
01What is context engineering
02System prompts and instructions
- Module 2: System prompts and instructions
- The system prompt as a budget item
- Where the system prompt goes: the top-level `system` parameter
- Rules for using the documents
- Handling missing information
- Output format instructions
- Auditing the system prompt: `lint_system_prompt`
- Mini-project: the Reservo Assistant's system prompt
03Structuring the context window
04Selecting content under a token budget
05Compaction (summarizing history and long results)
- Module 5: Compaction (summarizing history and long results)
- Why history grows and what to compact
- Extracting key fields into a structured record
- Summarizing a long result
- Two-stage compaction for long sessions
- What compaction must preserve
- Going beyond agent-fundamentals M6
- Mini-project: a compaction pass for the Assistant
06Few-shot examples as a budget line
07Context rot, isolation, and measuring context changes
- Module 7: Context rot, isolation, and measuring context changes
- What is context rot
- Simulating degradation as context grows
- Measuring a form signal with and without the techniques
- Isolation: giving a sub-task its own clean window
- The habit of measuring before and after
- Putting the measurement loop together
- Mini-project: measure a context change
08Reservo's support context pipeline (project)
Reviews
What students say
These reviews are from enrolled students who completed at least 50% of the course. We moderate reviews only on content grounds (spam, offensive language, personal data), never for being critical or negative.
No approved reviews yet.
Be the first to share your experience!