Factify
AI-powered news platform that aggregates European sources and synthesises unbiased stories using LLMs.
Overview
Factify started as a 48-hour hackathon project and evolved into my Business Analytics thesis at ICAI, scoring 94%. It’s a digital news platform that uses LLMs to compare and synthesise news stories from multiple European sources, cutting through editorial bias.
How It Works
- Source aggregation: crawls multiple European outlets for articles covering the same event
- Bias detection: identifies editorial slant, missing context and factual claims across sources
- LLM synthesis: generates a unified story presenting all perspectives with sourced claims
- Confidence scoring: rates each claim based on cross-source verification
Financial Analysis (Thesis)
The B.B.A thesis extended Factify into a full startup financial analysis:
- Competitor analysis: scatter plots comparing subscriber counts vs. pricing for Spanish newspapers
- Cost projections: team hiring costs, LLM API pricing comparisons and tokenizer/cost estimators
- Revenue modelling: Monte Carlo simulations with modifiable parameters for startup validation
- User surveys: validation data for problem/need understanding
- 3-year projections: income statements and financial forecasts
The financial models were built as interactive Dash applications in Python.
Technical Stack
Built in Python with Jupyter notebooks for rapid iteration. The NLP pipeline used pre-trained language models for summarisation and sentiment analysis. The Dash framework powered interactive visualisations for the financial analysis component.
What I Learned
The hardest problems in AI for journalism aren’t technical. What does “unbiased” actually mean? How do you handle conflicting facts? When should you surface uncertainty instead of a clean answer? Building the product and stress-testing the business model together gave a much fuller picture than either alone.