I used AI to distill 5,582 tweets from an X blogger into a reusable investment analysis Skill.

Origin

There's a trader on X (formerly Twitter) called Serenity (@aleabitoreddit) — 450K followers, focused on AI/semiconductor supply-chain analysis. His tweets are exceptionally high quality: he traces hyperscaler capex (Google, Microsoft, Meta, Amazon) upstream to find single-point bottlenecks, then positions before institutions react.

His methodology is clear: Don't buy NVDA. Walk the supply chain upstream and find the one chokepoint.

The problem: 5,582 tweets spanning 10 months, extremely high information density but wildly fragmented. Reading them all manually would take days, and key insights are scattered across replies and threads.

I thought: can I use AI to distill all these tweets into a structured, reusable analytical framework?

Method

1. Data Collection

Used agent-reach's twitter-cli with date-windowed search to work around X's pagination limits. Collected every tweet from 2025-07-02 to 2026-05-28 (~5,582 tweets) plus 4 long-form X Articles.

2. Period-by-Period Distillation

Split the tweets into 6 periods, ran a deep analysis on each with AI:

• 2025 Q3 (Jul-Sep): Early days, ALAB/CRDO/NBIS
• 2025 Q4 (Oct-Nov): NeoCloud explosion, NBIS becomes defining winner
• 2025-12 to 2026-01: Photonics awakening, LITE/AXTI/AAOI
• 2026-02: AAOI earnings blowout, SIVE emerges
• 2026-03: SIVE enters, CPO supercycle begins
• 2026-04-05: Supply-chain evidence chain closes

3. Cross-Merge

After all 6 period analyses, three consolidation passes:

• Per-ticker knowledge base: Merge each ticker across periods, tag conviction evolution
• Methodology extraction: Abstract 12 transferable principles from concrete cases
• Track record calibration: Independently verify 30/60-day accuracy using Yahoo Finance data

4. Package as Agent Skill

Final output: a standard Agent Skill that any AI agent (Claude, GPT, Cursor, OpenClaw) can use directly.

Core Methodology: 12 Transferable Principles

1. Bottleneck Hunting — Find the sole single-point-of-failure in a growing chain
2. Multi-Hop BOM Mapping — Chain from end-demand down to raw materials
3. Signed ARR vs. Market Cap Mismatch — Price on contracted forward revenue, not trailing multiples
4. Mag7 Customer Concentration Filter — Multiple Mag7 buyers = demand moat
5. The GAAP Margin War — Only trust GAAP margins, never cherry-picked non-GAAP
6. Qualification Cycle vs. TTM Revenue — Enter during qualification, before revenue confirms
7. Dilution / ATM Calendar as Disqualifier — Large ATM = structural ceiling
8. Counterparty / Financing Quality Spectrum — Who backstops the contracts determines survival
9. Short-Squeeze Setup (Profitable Grower) — High short interest + earnings growth = squeeze
10. Tariff / Macro Shock as Buy — Panic selling = best entry
11. Institutional Lag — Retail cognition leads institutional capital
12. Vega / IV Mispricing — Inefficiencies hiding in options structures

Test Run: Analyzing China's Thermoelectric Industry

After installing, I ran the Skill's 14-point checklist against China's thermoelectric (热电) industry. The results were revealing:

• Cogeneration plants (Dalian Thermoelectric, Hangzhou Thermoelectric) = regulated utilities, stable but no structural upside. Not Serenity's game.
• Thermoelectric materials (Bismuth Telluride, Bi₂Te₃) = Bi/Te under export controls, smelting byprouct supply is rigid, AI cooling demand exploding — this is the bottleneck logic.

Using his supply-chain mapping method, I traced the full chain:

AI Data Centers → Thermoelectric Cooling Modules → Bi₂Te₃ Crystals → Bi/Te Raw Materials → Zinc/Lead Smelting Byproducts

The real bottleneck isn't the raw materials (though they're also constrained) — it's high-purity Bi₂Te₃ crystal growth. Same logic as Serenity's InP substrate thesis.

Track Record Calibration (Independent Verification)

Never blindly trust any analyst. I independently verified Serenity's dated public calls using Yahoo Finance adjusted-close data:

Metric	Result
30-day directional accuracy	61% (30/49)
30-day ±10% strict hit	41% (20/49)
60-day +20% favorable close	54% (29/54)
Mature thesis validation rate	65-75%
AI photonics/CPO/InP/memory subset	75-85%

Verdict: His supply-chain bottleneck analysis deserves high reference weight. But specific trade operations (options, leverage, sizing) should never be copied.

How to Use

One command to install:

npx skills add WOOK98/serenity-aleabitoreddit

Then in any AI agent, when you ask about AI/semiconductor/photonics/memory/power stocks, it automatically triggers Serenity's analytical lens. Three workflows: evaluate a single ticker, audit a portfolio, or form a forward sector view.

Open Source

GitHub: https://github.com/WOOK98/serenity-aleabitoreddit

Fully public. 14 files, 4,226 lines. Includes the complete per-ticker knowledge base, methodology, track record with calibration, and 6 period analysis provenance files.

Final Thought

This experiment validated one thing: fragmented public information, when structurally distilled, becomes a reusable analytical tool.

X is full of high-quality analysis tweets, but they scatter across timelines and get forgotten. Distilling them into an Agent Skill is like "open-sourcing" an analyst's methodology.

If you follow any analyst or creator, you can use the same approach to distill their public views into an Agent Skill. The tooling is generic — what matters is finding information sources worth distilling.

---

Disclaimer: This post shares methodology and tooling only. It is not financial advice. Serenity's self-reported returns are unverified and carry survivorship bias. Investing involves risk; all decisions are your own.