InsightForge

Data-driven design, before the campaign begins - Shift ad campaigns from reactive ad optimization to proactive data-driven approach using customer voice, garnering customer engagement from day 1

Comment

InsightForge

Inspiration

-commerce sellers are flying blind. They spend massive budgets on marketing and ad campaigns using generic product photos that completely ignore what their customers actually care about, and what drives engagement. They take photos highlighting product features that they 'think' the customers would be attracted to, instead of monitoring customer reviews and comments. They reactively try to identify the problem when their campaign doesn't work. By then, they have lost time, resources and potential customer base. It's slow, expensive, and driven entirely by gut feeling.

How It’s Built

Stack: React/Vite frontend, FastAPI backend. Data: Selenium and BeautifulSoup4 for automated Amazon scraping. AI: KeyBERT for NLP, Gemini AI for synthesis, and Presaige AI for predictive scoring and creative optimization.

Technical Flow

  1. Ingest: Scrapes product data and reviews via URL.
  2. Analyze: Extracts VoC "pros and cons" using NLP.
  3. Score: Presaige API generates a Readiness Score to identify engagement gaps.
  4. Synthesize: Gemini AI generates a new ad based on VoC and Presaige insights.
  5. Refine: Presaige Recommendations API post-processes the image for maximum conversion.

Challenges & Learning

Presaige API Integration: Integrating the Presaige scoring API into our automated pipeline was non-trivial. We had to build a feedback loop where generated images are automatically scored and the highest-impact visual is selected without any human intervention. Getting this loop to work seamlessly end-to-end took significant engineering effort. Semantic Meaning vs Raw Keywords: Raw keywords like "battery life" don't tell you the strategic story. We overcame this by integrating KeyBERT with embedding-based semantic clustering so the system surfaces deeper themes like "Reliability is your critical weakness" rather than just listing raw words. Building End-to-End in Hackathon Time Connecting scraping → NLP → insight extraction → image generation → scoring into one seamless pipeline under time pressure was extremely challenging. We overcame this through a modular architecture where each component could be built, tested, and plugged in independently.

What’s Next

InsightForge is just getting started. Our immediate next step is launching a competitor analysis engine so instead of just analyzing one product, sellers can benchmark themselves against their top competitors automatically. One of our most exciting upcoming features is a Chrome Extension so users can run InsightForge on any product URL, on any eCommerce platform, directly from their browser. Just click, analyze, and get instant product intelligence while you browse.

Built With

  • beautifulsoup4-ai-integrations:-presaige-api-(predictive-ad-scoring-&-creative-recommendations)
  • fastapi
  • frontend:-react
  • google-gemini-ai-(image-generation-&-text-synthesis)
  • uvicorn-web-scraping-engine:-selenium
  • vite-backend:-python
Share this project:

Updates