Data Flow

1. User inputs product name, audience, and tone.
2. Agent 1 generates 3 ad variants.
3. Agent 2 predicts CTR for each.
4. Agent 3 reallocates spend accordingly.
5. Dashboard updates in real time.

---

5. Tech Stack

---

6. Task Breakdown

Total: 2.5 hours

---

7. Stretch Goals

• 🔁 Feedback loop: use predicted CTRs to refine ad generation prompts
• 📈 Visualize budget changes over iterations
• 🔗 Connect to Google Ads or Meta APIs for real ad testing
• 🧩 Integrate Weave for agent tracing and observability

---

8. Demo Scenario

Input:

Product: “AI Resume Builder for Job Seekers”
Audience: “Young professionals switching careers”

Output:

• 3 generated ad variants
• Predicted CTRs (e.g., 3.1%, 2.6%, 4.5%)
• Reallocated budget (e.g., $450 → best ad, $300 → next, $250 → lowest)
• Dashboard showing rankings and spend distribution bar chart

---

9. Naming, Tagline & Elevator Pitch

Product Name: AdSymphony
Tagline: “Where creativity meets optimization — in perfect harmony.”
One-liner:

AdSymphony is an AI-driven marketing orchestrator that automates ad generation, evaluation, and spend optimization through intelligent agents — saving marketers hours and maximizing ROI.

---

10. Repository Structure

/adsymphony ├── agents/ │ ├── copywriter.py │ ├── predictor.py │ └── allocator.py ├── app.py ├── data/ │ └── spend_simulation.csv ├── frontend/ │ └── streamlit_app.py └── README.md

---

API Endpoints

Calling POST /api/ads/generate

Example payload:

{
  "product_name": "AI Resume Builder for Job Seekers",
  "audience": "Young professionals switching careers",
  "tone": "Professional",
  "key_benefits": ["ATS-friendly formats", "Fast feedback"],
  "call_to_action": "Get started",
  "length_preference": "medium"
}

Response summary:

• variants – array of 3 ads with variant_id, headline, body, cta.
• metadata – request id, timestamps, token usage, model details.

Start the server with:

uvicorn app:app --reload --port 8000

Swagger UI: http://127.0.0.1:8000/docs
ReDoc: http://127.0.0.1:8000/redoc

---

11. Judging Fit

---

Authors:
Team of 4 — AI engineer, frontend dev, ML dev, and orchestrator.
Build Time: 2.5 hours
Hackathon Track: Self-Improving Agents
Sponsors: Weave, Daytona, Serverless RL, AG-UI, Tavily, Mastra