AI Project 2 · May 2026 · 11 Days

AI Content Engine

A system that synthesizes a curated reading list into intelligence records, then generates Tier 1/2/3 content — full articles, blog posts, and LinkedIn reactions — in your voice.

Claude API Google Firestore Google Sheets API Google Imagen 3 Python Streamlit Web Search
View on GitHub ↗ ← Back to site
How It Works
📋
Reading List
Google Sheets → bulk synthesize URLs
🔬
Synthesis
Claude extracts TLDR, key points, tier, themes
📚
Library
Firestore-backed intelligence database
✍️
Write
Web research + Claude generates tiered drafts
🚀
Publish
HTML export + LinkedIn content pack
Features
📰
Daily Intelligence Brief
Claude searches the web daily based on your auto-derived interest profile from your library. Finds 8-10 fresh articles. One-click synthesize any or all.
AGENTIC
🧠
Web-Researched Drafts
T1 and T2 generation searches 4-6 authoritative sources before writing. Every claim is cited. Output is publishable with your professional reputation behind it.
WEB SEARCH
📦
LinkedIn Content Pack
One article → five LinkedIn posts with genuinely different angles: Board Case, Practitioner, Urgency, Contrarian, Story. Each with image prompts and posting schedule.
5 ANGLES
🎨
AI Image Generation
Claude reads your draft and suggests 3 contextual image concepts. Google Imagen 3 generates blog headers (16:9) and LinkedIn cards (1:1) with one click.
IMAGEN 3
⬇️
HTML Export
T1 articles export as complete, styled HTML pages — dark/gold theme — ready to drop into any static site. Header image auto-embeds.
PUBLISH-READY
💾
Save Draft for Later
Right topic, wrong timing? Save any draft to Firestore. Returns when you're ready. Library tracks: synthesized → drafting → draft saved → published.
FIRESTORE
🔍
Cross-Article Synthesis
Search tab → pick a theme → synthesize N sources into one definitive take. How T3 articles become T1: accumulate evidence, then synthesize.
MULTI-SOURCE
🔗
Google Sheets Integration
Your reading list lives in Google Sheets. The engine reads it, synthesizes in batches, writes back status per row: Done / Duplicate / Skipped / Pending Paste.
GOOGLE API
Tier Classification
Claude classifies every article: T1 landmark sources, T2 standalone posts, T3 reactions. Auto-deduplication across 50+ articles via URL normalization.
HAIKU + SONNET
Content Tiers
Tier 1 — Full Article
1200-1800 words · usmanc.com
"Searches 4-6 authoritative sources. Full inline citations. Published as HTML with blog header image. The definitive CISO take."
Tier 2 — Blog Post
500-800 words · LinkedIn article
"Searches 2-3 sources. One focused angle. Cited claims. Sources section. Publishable standalone on LinkedIn or usmanc.com."
Tier 3 — Reaction
150-300 words · LinkedIn post
"Top 3 CISO takeaways. Emoji bullets. Named concepts. 'Link to original in comments.' Done in under 2 minutes."
Tech Stack
Anthropic Claude API (Sonnet + Haiku) Web Search Tool Google Firestore Google Sheets API Google Imagen 3 (AI Studio) Python 3.9 Streamlit BeautifulSoup (scraping) Firebase Admin SDK URL normalization (dedup) GitHub Pages (this site) Streamlit Cloud (app)