Evidence Media

In spring 2025, I launched Evidence Media, an independent news platform that automates the curation and publication of sourced news across X and Substack. What started as a personal project to counter mainstream media bias has evolved through five major versions into a fully autonomous media outlet covering 8 categories daily across multiple publication formats and platforms.

Evidence Media is a fully autonomous, AI-powered news pipeline designed to curate, analyze, and distribute independent news for North American audiences. It publishes daily across X (news briefs, threads, ultra-shorts, quote-tweets) and Substack (deep-dive articles by category, a free Daily Brief newsletter, and Notes throughout the day).

Since the COVID-19 pandemic, public trust in mainstream media has been severely undermined due to widespread collusion between institutions, governments, and news organizations. This convergence gave rise to a homogenized narrative, accompanied by systemic censorship—now widely referred to as the censorship industrial complex. In response, a growing number of citizens have turned to independent sources of information: freelance journalists, niche newsletters, alternative podcasts, and social platforms like X.

But in this decentralized landscape, staying genuinely informed has become increasingly time-consuming. Cross-referencing, verifying, and filtering a rising flood of scattered information is a task few people have time for.

That’s where Evidence Media comes in. Its mission is to filter and aggregate the most relevant and impactful independent news for North American audiences (Canada and the U.S.). The platform covers eight key categories: Business & Economics, Current Affairs, Environment, Health, International Affairs, Politics, Science & Technology, and Society.

Every piece of information shared is accompanied by its original source, allowing readers to assess its relevance, verify its accuracy, or explore further if desired. Unlike mainstream outlets bound by top-down editorial lines—or some independent sources that lack rigorous sourcing—Evidence Media is built on transparency, reliability, and methodological integrity.

The project launched in spring 2025. After four major iterations, V5 shipped in spring 2026 as a complete ground-up rebuild with comprehensive test coverage, automated workflows, structured logging, and a continuous editorial improvement system.

Evidence Media was a fully solo project. I handled every layer from vision to execution:

• Product Manager: I defined the mission, editorial identity, positioning, target personas, and long-term vision.
• Backend Development: I built the entire V5 architecture in Python following TDD/SOLID principles, with domain models, adapters, pipelines, and 919 unit tests.
• AI Integration: I orchestrated a multi-model pipeline using xAI Grok (all LLM tasks), OpenAI (embeddings), and Anthropic Claude (self-healing Selenium), with batch API optimization and prompt caching for cost efficiency.
• Advanced Automation: I designed a fully automated CI/CD pipeline with early-exit guardrails, dual failure alerts (issue tracking + email), and intelligent scheduling across multiple daily publishing windows.

From concept to iteration, from the first line of code to editorial strategy, I drove every decision with one clear goal: to build a resilient, automated, and trustworthy news platform that serves the public good.

The creation of Evidence Media was born from a clear realization: staying properly informed has become both complex and time-consuming. This growing information gap allows politicians, institutions, and influential actors to operate without real checks and balances, a serious threat to any democratic society. Where journalism once served as a safeguard, traditional media is now largely owned by the very entities it should be holding accountable, breaking the bond of public trust.

Evidence Media is a direct response to this imbalance, built on three core pillars: a personal observation, a tangible need for clarity in a fragmented media landscape, and a creative drive to structure independent information in a way that is both accessible and credible.

My studies at HEC Montréal shaped the project's strategic vision and business model, while my hands-on experience with Wise Duck Dev GPTs and Jean The Writer equipped me with the expertise in artificial intelligence and automation needed to develop a solution that is robust, reliable, and fully scalable.

Evidence Media was born from a clear mission: to deliver independent, sourced, reliable, and relevant information, with maximum signal and minimal noise. I began with a functional MVP (Version 1) focused on automation, content quality, and editorial integrity.

Rather than building around predefined personas or chasing algorithmic trends, I created the media outlet I wished already existed—one aligned with my core values, free from the noise of traditional news cycles, and respectful of the reader’s intelligence.

This product-centric and values-driven approach naturally led to the development of a unique political doctrine, which became the project’s editorial backbone—ensuring coherence, honesty, and civic responsibility across every piece of published content.

V5 publishes multiple content formats on X throughout the day across strategic peak windows, including news briefs, threads, ultra-shorts, and quote-tweets. On Substack, it publishes daily articles by category (paid), a free Daily Brief newsletter every morning, and Notes throughout the day.

Growth strategy focuses on content quality over vanity metrics. The free Daily Brief builds trust and audience, paid articles convert engaged readers, and a Kaizen ideology feedback loop continuously improves the editorial framework based on real-world story coverage.

The methodology was agile and iterative—constantly refining the formats, editorial tone, and automation workflows based on personal observations, user feedback, comments, and algorithmic signals.

Finally, the system was designed for resilience and scale: 100% automated, agent-ready, and structured to evolve alongside the capabilities of the underlying APIs.

• Language: Python 3.12
• AI: xAI Grok (all LLM tasks), OpenAI (embeddings), Anthropic Claude (self-healing)
• APIs: xAI SDK, xdk (X official SDK), OpenAI API, Anthropic API, Google API, Wikidata API
• Automation: GitHub Actions (CI/CD), cron scheduling, intelligent daily randomization
• Security: HashiCorp Vault, OAuth 2.0 token rotation
• Publishing: curl_cffi (Substack API), Selenium (fallback), Tweepy (X media upload)
• Quality: pytest (comprehensive test suite), Pydantic (domain models), GCP-compatible structured logging

The V5 architecture follows clean separation of concerns: domain models (Pydantic), adapters (external APIs), pipelines (orchestration), and entrypoints (CLI). Every external integration is wrapped in a typed adapter with retry logic, structured logging, and cost tracking.

Cost optimization is central: xAI Batch API (50% token discount), prompt caching via conversation IDs (up to 75% off), and centralized pricing tracking across all providers.

Evidence Media doesn’t have a traditional UI, its user experience is entirely content-driven. I focused on crafting a readable, reliable, and recognizable content format, inspired by the most effective X and Substack accounts. Every post is designed to deliver immediate clarity and traceable credibility by citing original sources, similar to academic footnotes, helping users assess news at a glance or explore further.

The main UX challenge was LLM hallucinations and inconsistency on politically sensitive topics. I addressed this through aggressive prompt engineering, and custom constraints, significantly reducing factual drift. Still, certain topics remain inherently unstable in current models, underscoring the importance of transparent sourcing.

To support discoverability and habit-building, I introduced consistent formatting across X posts (short, sourced, high-signal) and Substack articles (longer, categorized, contextualized), reinforcing trust and boosting engagement over time.

Evidence Media runs entirely from a GitHub-hosted codebase using GitHub Actions for scheduled, event-driven automation. While it doesn't rely on traditional deployment platforms (like Vercel or AWS), its architecture is optimized for resilience, modularity, and infinite scalability.

Each functional component, from scraping to AI generation to publishing, is encapsulated in discrete scripts that can be scaled horizontally or triggered independently, as needed.

The system supports CI/CD via GitHub Actions, while monitoring and performance analytics are handled through native platform dashboards (X and Substack). The only true limitations are API usage quotas (xAI, X API, OpenAI, etc.), which define throughput, but the architecture itself is capable of 24/7 continuous publishing at industrial scale with minimal adjustments.

In short, the pipeline is not just automated, it’s built to grow.

• V1: X pipeline (short-form, sourced posts).
• V2: Content redesign, improved cadence and formatting.
• V3: Substack integration (daily sourced articles).
• V4: Fully autonomous 24/7 pipeline on Raspberry Pi + GitHub Actions.
• V5 (current): Complete ground-up rebuild with clean architecture, multiple publication formats, self-healing infrastructure, paid tier, and continuous editorial improvement.

Next: fine-tune a model on the curated dataset for improved cross-story context and historical linking.

Evidence Media's X account grew to over 500 organic followers, with consistent growth driven by sourced, high-quality content. Substack adoption is growing steadily with the introduction of the paid tier and Daily Brief. Audience feedback validates the editorial approach: readers appreciate the transparency, consistent sourcing, and the publication's willingness to cover stories mainstream outlets ignore.

Engagement varies by topic, but the overall reception validates the project’s mission, offering verifiable, independent information in a time of institutional distrust. The system’s success further confirmed that automation, sourcing, and fast iteration are key levers for building trust and reach in digital media.

Evidence Media wasn’t just another automation project, it was a deep dive into building a living, breathing AI-powered media outlet from scratch. It sharpened every dimension of my technical skill set: Python, web scraping, prompt engineering, data integrity, CI/CD pipelines, API orchestration, and cybersecurity with Vault. I built systems that are not only fast and scalable, but also resilient, verifiable, and transparent, essential traits when working in information distribution.

I learned to tame large language models in high-stakes editorial contexts, resolving hallucinations, bias, and inconsistency through layered prompt strategies and dynamic content filtering. I developed ways to ensure that AI supports human understanding rather than distorting it, preserving truth and traceability through academic-style sourcing.

But beyond the tech, this project solidified a core truth: building something truly useful doesn’t come from chasing trends or audience metrics, it comes from solving your own problem first, at scale. I built the news outlet I was searching for. The one I needed, but couldn’t find.

Most importantly, this project reaffirmed a principle I now apply everywhere: quick, thoughtful iteration beats perfection. Shipping fast, observing real-world feedback, and improving continuously is the fastest path to building reliable, high-impact systems.

Key takeaway: Think independently. Iterate rapidly. Automate relentlessly. Build solutions that you would genuinely use—and others will follow.