Artificial Intelligence
Turning unstructured data into actionable insights - at scale, and at speed
Company name under NDA
Role: Product Manager (AI Focus)
Product: AI-Driven Sentiment Analysis Platform
Timeline: 2024-2025
Impact: Automated manual news and social media sentiment analysis for political campaign teams—cutting research time from hours to minutes and enabling faster, insight-driven decision-making.
Context
Political consultants and campaign teams were manually reviewing news and social media to gauge voter sentiment in targeted regions—a labor-intensive process that limited their ability to react quickly and reliably to changing public opinion.
Issue:
The team was taking hours reviewing articles to extract sentiment data before any analysis could take place.
Approach
AI Automation for Human Insight:
- Designed a process to automate the ingestion, summarization, and sentiment analysis of news and (soon) X.com posts using the OpenAI API.
- Collaborated with the client to model the manual workflow and define the standards for "insight similarity."
Process Mapping & Modularization:
- Built a pipeline: Web scraper → OpenAI summary generator → OpenAI sentiment analysis → Automated report output (PDF).
- Ensured outputs were easy to review and actionable for campaign strategists.
Iterative Benchmarking:
- Compared AI-generated insights against manual results, refining prompts and evaluation logic based on client feedback.
Execution
- Developed a modular pipeline (see diagram above):
- Web Scraper: Automated article/content extraction from curated lists.
- OpenAI Summary Generation: Used LLMs to create concise summaries from raw content.
- Sentiment Analysis: Applied OpenAI sentiment tools to summaries for clear, region-tagged outputs.
- Report Generation: Assembled results into a standardized, client-ready PDF for immediate use.
- Ran side-by-side pilot tests to measure speed and "similarity of insight."
- Collected feedback and iteratively tuned the pipeline for accuracy and utility.
Outcome
- Reduced research time from 2–3 hours per 6-article batch to under 2 minutes.
- Delivered insights judged by clients to be ~90% similar to manual, human-generated sentiment (subjectively measured).
- Freed up client teams to focus on interpreting and acting on insights—not gathering or wrangling data.
- Roadmap established for integrating X.com (Twitter) API for live social sentiment coverage.
Tools & Frameworks Used
- Python (ETL pipeline, automation)
- OpenAI API (summarization & sentiment analysis)
- Web scraping libraries
- JSON for data exchange
- PDF generation (report output)
Capabilities Demonstrated
- Primary: Artificial Intelligence (AI)
- Secondary: Automation, NPD, Data Visualization
Key Learnings
- Human judgment of “accuracy” in sentiment is best measured by similarity, not precision—requiring both technical and collaborative tuning.
- Modular, API-driven processes enable rapid iteration and future expansion (e.g., adding new data sources).
- Clear, visually digestible outputs (PDF reports) drive real adoption in client workflows.
