LLM–Marketing Hybrid: Adapters + Rewards (Repo Skeleton)

This repository contains:

• Paper templates (LaTeX and Markdown)
• Experiment framework with synthetic-data scripts to reproduce the figures for:
  • E1: Compute scaling curves
  • E2: Domain-knowledge ablation
  • E3: Cross-channel transfer via adapters
  • E4: Offline→Online correlation
  • E5: Safety/compliance under search
• Mind-map JSON export for your website

Core idea: Follow Sutton’s “bitter lesson.” Keep domain specifics as adapters + rewards and put performance in general methods (search, learning, retrieval).

Quickstart

# 1) Create a virtual env (optional)
python3 -m venv .venv && source .venv/bin/activate

# 2) Install dependencies
pip install -r requirements.txt

# 3) Run all experiments
python src/experiments/run_all.py

# 4) See outputs
ls outputs/

Outputs are saved to outputs/ as PNG figures and CSVs. You can tweak experiment knobs in each script.

Repo Layout

paper/
  main.tex
  references.bib
  paper.md
mindmap/
  llm_marketing_mindmap.json
src/
  adapters/
    schemas.py
    connectors/
      ga4.py gads.py dv360.py crm.py
  retrieval/ (index + rag stubs)
  planner/ (search + evolution stubs)
  policy/ (bandit, offline replay, budget RL stubs)
  sim/ (click/conv simulator)
  eval/ (CUPED, sequential testing, metrics)
  actuators/ (API writers)
  governance/ (guardrails + versioning)
  experiments/ (E1–E5 + run_all)
  utils/ (logging + config)
outputs/
requirements.txt
.gitignore
LICENSE
Makefile

Notes

• All experiment scripts run with synthetic data to validate methodology, plots, and logging.
• Replace stubs in adapters/, actuators/, and governance/ with real platform code and policies.
• Figures are referenced in paper/main.tex and paper.md (update paths if needed).