Skip to content

Implemented All Course Run in FMS#2

Open
AmandaHalim wants to merge 2 commits into
mainfrom
Amanda
Open

Implemented All Course Run in FMS#2
AmandaHalim wants to merge 2 commits into
mainfrom
Amanda

Conversation

@AmandaHalim
Copy link
Copy Markdown
Collaborator

No description provided.

Afzaana512 added a commit that referenced this pull request May 6, 2026
… phase 2

Three fixes addressing user-reported issues in deck g3:

1. Strip [Sub-topic] markers from fallback bullets:
   The orchestrator collapse step inserts inline [Sub-topic name]
   markers in the bullet stream so Phase 2 AI knows where sub-topic
   boundaries are. The fallback was treating those markers as real
   bullets, producing slides titled "[What are Chatbots?]" with cards
   showing the bracket text. Now filtered before the fallback loop.

2. Action-verb labels in Block 0 overview + Key Takeaways:
   Previously block-0 cards showed "Definition and overview" as label
   and "Definition and overview of chatbots" as desc — visually
   redundant. Now cards rotate action-verb labels (Define / Apply /
   Identify / Analyse / Implement / Compare / Evaluate / Master /
   Practice / Review) with matching distinct icons (target/cog/
   lightbulb/chart-line/rocket-launch/scale-balance/check-circle/
   star/handshake/file-search) and the full bullet as desc. Same
   pattern in Key Takeaways with Remember/Practise/Apply/Reflect/
   Master verbs. Visually mirrors the Streamlit reference deck where
   block 0 shows "Define Goals" / "Identify Uses" / "Select Platform"
   with distinct icons.

3. Aggressive 4-attempt retry for Phase 2 (was 2):
   User requirement: "no fallback at all". Each attempt simplifies
   the request more aggressively to maximise AI success in production:
     #1: Full prompt + research enrichment + WebSearch (if sources<2)
     #2: Same prompt minus research details (smaller payload)
     #3: Tiny prompt — topic + bullets only, no research, no tools
     #4: Tiny prompt + Sonnet 4.6 (different model in case Haiku 4.5
         has a transient issue with this token/account)
   Only after ALL 4 fail does fallback run. Each attempt logs its
   model + prompt size + tools so production logs can show which
   tier succeeded (or all failed).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant