Inspiration
📘 Project Story: PharmaPulse AI
🌱 Inspiration
The idea for PharmaPulse AI came from observing how difficult it is to understand how a medicine is actually perceived beyond clinical trials.
Information about drugs is scattered across FDA documents, research papers, and online discussions, written in complex language that is not easily accessible to everyone.
During the hackathon, when we read the problem statement “Analyze web, social, and literature data to understand how healthcare professionals and patients perceive pharmaceutical products”, it immediately clicked — perception is missing from most pharma analytics tools. That gap inspired us to build PharmaPulse AI.
🎯 What We Built
PharmaPulse AI is a browser-based, AI-powered platform that:
- Retrieves regulatory and literature-level information
- Analyzes web-based public perception
- Uses local AI to summarize insights in a simple, human-readable format
The output is structured into three clear sections:
- FDA / Literature View
- Web & Public Perception
- AI Medicinal Summary (uses, simplified understanding, and price range)
Our focus was not diagnosis or advice, but perception intelligence.
🧠 What We Learned
This project helped us learn and apply:
- How unstructured medical text differs from general data
- The importance of prompt control to avoid irrelevant or unsafe AI outputs
- How to use local large language models when paid APIs are not an option
- Building a full-stack application using Flask, HTML/CSS, and JavaScript
- Designing AI systems that summarize, not advise
We also learned that clarity > complexity when presenting healthcare data.
🛠️ How We Built It
🔹 Architecture Overview
- Backend: Python + Flask
- Frontend: HTML, CSS, JavaScript (browser-based UI)
- AI Engine: Local LLaMA model via Ollama
- Data Sources: Regulatory literature + web perception
The AI acts as a controlled summarizer, not a medical expert.
🔹 Core Idea (Conceptual Flow)
[ \text{Perception Insight} = f(\text{FDA Data} + \text{Web Signals}) ]
The AI function ( f(\cdot) ) condenses multiple sources into a short, meaningful summary.
⚠️ Challenges We Faced
- API limitations: Many powerful AI APIs are paid, so we had to shift to a local model
- Irrelevant results: Early outputs included non-medicinal data (places, chemistry, labs)
- Over-safety responses: The AI initially refused to summarize due to medical disclaimers
- Environment issues: Python, virtual environments, and Windows setup required careful handling
Each challenge forced us to refine prompts, architecture, and logic, resulting in a more robust system.
🚀 Final Outcome
PharmaPulse AI demonstrates how AI can responsibly bridge the gap between complex medical data and real-world understanding.
It empowers users with perception-level insights — fast, clear, and accessible.
This project represents not just a tool, but a **direction for future pharma intelligence platforms
Built With
- a-local-llama-model-via-ollama-for-ai-insights
- css
- fda-opendata-api
- flask
- html
- html/css/javascript-for-the-frontend
- javascript
- ollama
- python