Trace a prompt end-to-end through LangChain Expression Language and produce the architecture diagram. The diagram identifies the Jinja2 firing point and classifies each component by its OWASP LLM or ASI risk category.
What you ship
- An architecture diagram (ASCII art or PNG) labeling each LangChain component.
- A one-page narrative walking the prompt path from user input to model API call.
- A short README naming the LangChain version pinned for the trace.
- Toolchain Diary entries for any new tools you touched.
Tools you use
- A clean Python virtualenv with the cohort-pinned LangChain.
- Burp Suite Community to intercept the model-API HTTP call.
- Optional: the academy pcap analyzer for the alternative trace path.
Success criteria
- The diagram shows the Jinja2 firing point explicitly.
- Each component carries an OWASP LLM or ASI classification.
- The narrative is precise enough that an instructor can rebuild the trace from your repo.
Time budget
Plan for two ninety-minute lab sessions plus two hours of independent build-out. Modules 4 and 6 commonly run over; budget one extra session for those.
Submission
Push to your student repo under adv-102/labs/lab-2/. Include source, a one-paragraph README, the output you observed, and where applicable a structured detector or trace file.