ainewsletterpythonfintecheditorial
September 2025 Active

AI Distilled & Decoded

Monthly AI intelligence brief distributed to Equifax leadership. Curated 60+ articles per edition, built a Python + Perplexity pipeline to fetch and score content, added ElevenLabs TTS narration. 6 editions produced covering AI in financial services, model governance, agentic commerce.

Built with

Python
Python

Context

By mid-2025, Equifax’s leadership team was drowning in AI news but starving for signal. Everyone had seen the ChatGPT announcement. Nobody had a clear view of what mattered for a data-and-decisioning company specifically — model governance, AI in credit risk, synthetic data regulation, agentic systems in financial services.

I built AI Distilled & Decoded to solve that.

What It Was

A monthly intelligence brief — hand-curated and AI-augmented — distributed directly to VP-and-above leadership. Not a link dump. Not a summary bot. A point-of-view document that said: here’s what happened this month, here’s what it means for us specifically, here’s what you need to know to lead informed conversations.

6 editions produced covering:

  • AI in financial services and credit decisioning
  • Model governance and EU AI Act implications
  • Agentic commerce and autonomous financial workflows
  • LLM evaluation frameworks for enterprise deployment
  • Synthetic data generation and regulatory posture
  • GPT-4o, Claude 3, and Gemini comparative analysis for business use cases

The Pipeline

01
Discover
60+ articles via Perplexity API, arXiv, Substacks, industry reports
02
Score
Python ranks by Relevance · Novelty · Signal strength
03
Synthesize
Claude drafts "so what for us" narrative — not summaries
04
Narrate
ElevenLabs TTS generates audio edition for commute listening

Content Discovery

  • 60+ articles reviewed per edition from Substack newsletters, arXiv pre-prints, industry reports, and major tech publications
  • Perplexity API for targeted search: "AI + financial services + [month]" queries with quality filtering

Scoring & Selection

  • Python script scores each article on 3 dimensions: Relevance (fintech/credit), Novelty (not already covered), Signal strength (leadership actionability)
  • Top 8–12 articles selected for the brief

Production

  • Claude drafts the narrative analysis — not summaries, but synthesis with explicit “so what for the business” framing
  • ElevenLabs TTS generates an audio version for commute listening
  • Distributed via email with both PDF and audio attachment

What I Learned

The hardest problem wasn’t content discovery — it was framing. A 300-word explanation of why the EU AI Act matters to a VP who manages 200 people and has 7 minutes to read is a completely different document than a journalist’s explainer.

The audio version turned out to be unexpectedly high-engagement. Several leaders mentioned they listened on their morning run. ElevenLabs’ quality cleared the bar for professional audio without sounding robotic — which meant the narration got treated as content rather than a gimmick.

← All Projects