Evidence Media

Automated, AI-powered news curation & publishing

In spring 2025, I launched Evidence Media—a fully automated pipeline that curates independent news and publishes sourced updates to X and longer, contextual articles to Substack. The goal: cut through noise, surface what matters, and make every claim traceable.

Project Overview

Evidence Media filters and aggregates high-signal stories across Business & Economics, Current Affairs, Environment, Health, International Affairs, Politics, Science & Tech, and Society. Each post cites original sources (think academic footnotes) so readers can verify quickly or dig deeper. The system is 100% automated end-to-end: discovery → analysis → drafting → sourcing → publishing.

My Role

Solo build, from vision to execution:

• Product & Editorial Strategy: mission, voice, cadence, categories, and KPIs.

• Backend & Automation: modular Python scripts orchestrated as a pipeline.

• AI Orchestration: multi-model stack (OpenAI, xAI, Perplexity) for context, summarization, and sourcing.

• Security & CI/CD: GitHub Actions for scheduling; secrets managed with HashiCorp Vault.

Stack & Tools

Python, Flask, BeautifulSoup, Selenium, Tweepy, GitHub Actions, HashiCorp Vault, plus OpenAI/xAI/Perplexity APIs, X API v2, and Google APIs. Chosen for reliability, strong docs, and automation-friendly ecosystems.

Design & UX Highlights

Content is the interface. I standardized concise, source-backed X posts and deeper, categorized Substack briefs. To reduce LLM hallucinations and bias, I used layered prompt strategies, constraints, and dynamic filtering—always prioritizing verifiability.

Deployment & Scalability

The pipeline runs from a GitHub-hosted codebase with scheduled workflows. Components (scraping, AI generation, formatting, publishing) are decoupled and horizontally scalable. Real limits are API quotas, not architecture.

Outcomes

• X account grew past 500 organic followers; Substack growth steady.

• Consistent engagement on sourced, high-signal topics.

• Stable, resilient pipeline capable of continuous publishing with minimal ops.

What I Took Away

• Deepened expertise in Python automation, scraping, prompt engineering, data integrity, CI/CD, API orchestration, and security.

• Learned to make AI helpful in editorial contexts—supporting human understanding with transparent sourcing.

• Reaffirmed a core product principle: iterate quickly, observe real-world feedback, improve continuously.

If you’re exploring AI-driven media or research automation, and want systems that are fast, scalable, and verifiable, I’d love to chat.