Upload 6 files

Dockerfile

@@ -0,0 +1,7 @@
+FROM python:3.10-slim
+WORKDIR /app
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+COPY . .
+# Using 4 workers to handle multiple concurrent API requests
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860", "--workers", "4"]

agent/core.py

@@ -0,0 +1,109 @@
+import os
+import json
+import random
+import base64
+from io import BytesIO
+from PIL import Image
+from groq import Groq
+from .personas import PERSONAS
+
+# The client will automatically use the GROQ_API_KEY environment variable
+client = Groq()
+
+def prep_image_bytes(image_bytes: bytes) -> str:
+    img = Image.open(BytesIO(image_bytes))
+    buffer = BytesIO()
+    img.save(buffer, format='PNG')
+    return f"data:image/png;base64,{base64.b64encode(buffer.getvalue()).decode()}"
+
+def get_image_description(image_data: str) -> str:
+    prompt = """<|begin_of_text|><|start_header_id|>user<|end_header_id|>
+Analyze this cosmetics ad image as Instagram post:
+• Product (serum/lipstick)?
+• Colors/lighting/vibe?
+• Text/CTAs?
+• Target audience/emotions?
+• Engagement hooks/weaknesses?
+Concise bullet summary:<|eot_id|><|start_header_id|>assistant<|end_header_id|>"""
+
+    chat = client.chat.completions.create(
+        model="meta-llama/llama-3.2-90b-vision-preview", # Updated to a stable Groq vision model
+        messages=[
+            {"role": "user", "content": [
+                {"type": "text", "text": prompt},
+                {"type": "image_url", "image_url": {"url": image_data}}
+            ]}
+        ],
+        temperature=0.3,
+        max_tokens=250
+    )
+    return chat.choices[0].message.content.strip()
+
+def simulate_user(image_desc: str, persona: dict) -> dict:
+    agent_prompt = f"""You are {persona['name']}, {persona['age']} {persona['gender']}, {persona['bias']}.
+Instagram ad scroll: {image_desc}
+**Think aloud** (1 sentence, your real reaction), THEN decide naturally:
+- Reaction: love/like/wow/haha/neutral/angry (text names)
+- Comment: AUTHENTIC Instagram - SHORT, slang/emojis/hashtags/varied! Cosmetics chat. null ok.
+- Click shop?: true/false
+Varied examples: "glowy af 🔥", "pricey sis?", "need this asap 💸✨", "#skincarejunkie"
+JSON ONLY: {{"thought": "private", "reaction": "love/etc", "comment": "varied text or null", "click": true/false}}"""
+
+    chat = client.chat.completions.create(
+        model="llama-3.1-8b-instant", # Fast, stable model for rapid agent simulation
+        messages=[{"role": "user", "content": agent_prompt}],
+        temperature=1.0,
+        top_p=0.9,
+        max_tokens=100
+    )
+
+    try:
+        data = json.loads(chat.choices[0].message.content)
+        reaction_map = {"❤️": "love", "💙": "like", "😍": "love", "👍": "like", "😂": "haha", "🤔": "wow", "😡": "angry"}
+        if data.get("reaction") in reaction_map:
+            data["reaction"] = reaction_map[data["reaction"]]
+        return data
+    except:
+        return {"reaction": "like", "comment": None, "click": False}
+
+def run_simulation(image_bytes: bytes, num_users: int = 100, impressions: int = 10000) -> dict:
+    image_data = prep_image_bytes(image_bytes)
+    desc = get_image_description(image_data)
+
+    engagements = {
+        "post_id": f"sim_{random.randint(10000, 99999)}",
+        "ad_analysis": desc,
+        "insights": {
+            "impressions": impressions,
+            "reach": int(impressions * 0.7),
+            "likes": 0
+        },
+        "reactions": {"love": 0, "like": 0, "wow": 0, "haha": 0, "sorry": 0, "angry": 0},
+        "comments": [],
+        "clicks": 0
+    }
+
+    users_sample = random.choices(PERSONAS, k=num_users)
+    for persona in users_sample:
+        user_data = simulate_user(desc, persona)
+
+        if user_data.get("reaction") in engagements["reactions"]:
+            engagements["reactions"][user_data["reaction"]] += random.randint(1, 5)
+            engagements["insights"]["likes"] += 1
+
+        if user_data.get("comment"):
+            engagements["comments"].append({
+                "id": f"c_{random.randint(1,999)}",
+                "message": user_data["comment"],
+                "like_count": random.randint(0, 20)
+            })
+
+        if user_data.get("click"):
+            engagements["clicks"] += random.randint(1, 10)
+
+    engagements["insights"]["ctr"] = round((engagements["clicks"] / impressions) * 100, 2)
+    engagements["insights"]["engagement_rate"] = round((
+        sum(engagements["reactions"].values()) + len(engagements["comments"])
+    ) / impressions * 100, 2)
+
+    return engagements

agent/personas.py

@@ -0,0 +1,7 @@
+PERSONAS = [
+    {"id": 1, "name": "GenZ_Influencer", "age": "18-24", "gender": "F", "bias": "trendy, emojis, hype", "engagement_bias": 0.8},
+    {"id": 2, "name": "Skeptical_Mom", "age": "30-45", "gender": "F", "bias": "practical, questions efficacy", "engagement_bias": 0.4},
+    {"id": 3, "name": "Luxury_Buyer", "age": "25-35", "gender": "F", "bias": "premium, aesthetics", "engagement_bias": 0.7},
+    {"id": 4, "name": "Casual_Shopper", "age": "20-30", "gender": "M", "bias": "simple, value", "engagement_bias": 0.5},
+    {"id": 5, "name": "Beauty_Nerd", "age": "22-28", "gender": "F", "bias": "ingredients, science", "engagement_bias": 0.6},
+]

main.py

@@ -0,0 +1,18 @@
+from fastapi import FastAPI, UploadFile, File, Form
+from agent.core import run_simulation
+
+app = FastAPI(title="Ad Simulation API")
+
+@app.get("/")
+def home():
+    return {"message": "Ad Agent is running in the cloud!"}
+
+@app.post("/simulate")
+async def simulate(
+    file: UploadFile = File(...),
+    num_users: int = Form(20),
+    impressions: int = Form(1000)
+):
+    contents = await file.read()
+    results = run_simulation(contents, num_users=num_users, impressions=impressions)
+    return results

requirements.txt

@@ -0,0 +1,6 @@
+fastapi
+uvicorn
+python-multipart
+groq
+pillow
+pandas