Skip to content

Releases: manojkp08/AdPulse

AdPulse MVP | Automated Insight Engine

03 Dec 08:27
@manojkp08 manojkp08
v1.0
df6ef29

Choose a tag to compare

View all tags
AdPulse MVP | Automated Insight Engine Latest
Latest

AdPulse MVP

LIVE APPLICATION

➡️ https://groundtruthaihackathon2025manoj.streamlit.app/

AdPulse removes the manual work from weekly reporting. It takes raw campaign logs (CSV) and instantly generates a PDF briefing with stats and AI-driven strategic context.

Key Features Delivered:

I focused on building a deployable, crash-proof tool.

  • Deterministic Logic: I used Pandas for all math (CPA, CTR, CPC, Impressions, Clicks, Spend, Conversions, Top Campain & ROAS). This ensures financial metrics are accurate every time.
  • AI for Strategy, Not Math: Accurate KPI calculations using Pandas .
  • AI Insights: Analysis generated via Google Gemini 2.5 Flash.
  • PDF Export Option PDF generation with visual metric cards.

Demo Video

AdPulse Demo

Screenshots

Note

Before testing, please review the CSV format guidelines — a fixed schema is required so the math engine can compute accurate KPIs. Details are provided below in the CSV File Parameters and Limitations (Inside 'How to Evaluate section') sections.

CSV File Parameters

Below are the key fields used from the uploaded CSV and how each metric is calculated:

  • Impressions → Total number of ad views.
    Formula: total_impressions = SUM(Impressions)
  • Clicks → Number of users who clicked the ad.
    Formula: total_clicks = SUM(Clicks)
  • Conversions → Successful outcomes (purchase, lead, install).
    Formula: total_conversions = SUM(Conversions)
  • Spend → Total ad spend in USD.
    Formula: total_spend = SUM(Spend)
  • CTR (Click-Through Rate) → Measures ad engagement.
    Formula: CTR = (Clicks / Impressions) * 100
  • CPA (Cost Per Acquisition) → Cost to generate a single conversion.
    Formula: CPA = Spend / Conversions
  • CPC (Cost Per Click) → Cost for each click.
    Formula: CPC = Spend / Clicks
  • Top Campaign → Highest conversion-generating campaign.
    Logic: idxmax() on Conversions
  • Best Platform → Platform with highest ROAS (Return on Ad Spend) proxy.
    Formula: ROAS_Proxy = Conversions / Spend

Workflow

image

How to Evaluate (Test Scenarios included):

I have included multiple datasets in the custom_csv/ folder to demonstrate the AI's adaptability:

Limitation ⚠️:

Please ensure your CSV contains the columns: Impressions, Clicks, Conversions, Spend, Campaign_Name, Platform. Otherwise analysis will fail because unstructured to structured conversion is not implemented in my MVP yet (Will be implemented in future), so a CSV file with this exact schema must be uploaded in the AdPulse platform. Check the Future Improvements & Scalability section for the roadmap to scale this beyond the current MVP limitations.

  • campaign_data.csv: Standard Mix.
  • adCampaign_sample2.csv: Larger dataset.
  • And many more CSV files to test.

Simply upload these to the app to see the narrative change dynamically.

Future Improvements & Scalability

Given more time beyond the 4-hour limit, here is how I would scale this into a full enterprise platform:

  • Direct Data Connectors : Currently, I used CSV uploads for the MVP. In production, I would add SQL/BigQuery connectors to fetch live campaign data automatically.
  • Unstructured Data Pipeline : I will implement an ETL layer to clean and structure messy data (e.g., random text reviews or emails) before feeding it into the calculation engine.
  • Format Flexibility : I would like to add support for .PPTX (PowerPoint) export using python-pptx for editable slides.
  • Multi-Domain Adapters : The current logic is optimized for AdTech. I would modularize the math engine to support other verticals like Retail (Sales/Inventory) or Logistics without changing the core code.

Assets 2
Loading