Director, Global AI/Copilot GTM Lead @ GitHub | Founder & CEO @ Entopiq, Inc

💻 GitHub — Leading global go-to-market for Copilot across AMER, EMEA, and APAC. Helping developers ship faster with AI.

🌱 Entopiq — Video-first talent discovery. Replacing résumés with 90-second videos, AI matching, and an AI Headhunter that finds candidates for you. Demo: Watch on X

🦞 Skarnfall — An agent-native talent marketplace where AI agents register, bid on tasks, and get paid. Agents are first-class users. The API is the product. Built with FastAPI, PostgreSQL, OAuth (Google/GitHub), encrypted payments (USDC, PayPal, etc.). skarnfall.com

✴️ How Boris Uses Claude Code — I've been deep in Claude Code since it launched. When Boris Cherny (the creator of CC) dropped a 13-part thread on how he actually uses it, I decided to memorialize it. Terminal aesthetic, tab navigation, built with the tool it's about. howborisusesclaudecode.com | X post

🔥 The "Just Fucking Use" Collection — For developers who need to stop overthinking:

• justfuckinguseclaudecode.com
• justfuckingusecopilot.com
• justfuckingusegemini.com
• justfuckingusenanobanana.com
• justfuckingusedeepseek.com
• justfuckingusefigma.com
• justfuckinguseperplexity.com
• justfuckingusesupabase.com

Fine-tuned Qwen2.5-3B-Instruct using LoRA + MLX on a base M4 Mac Mini (16GB). No cloud, no rented GPUs.

The model gives direct, opinionated career advice — no "it depends" hedging. Trained on 367 examples covering salary negotiation, counteroffers, burnout, IB/PE/VC transitions, and more.

Stats: 38 min training time · 7.19 GB peak memory · 15 tokens/sec inference · $499 hardware

📝 X Article: I Fine-Tuned Qwen on a $499 Mac Mini. Here's What Actually Worked.

🔗 Repo: qwen-fine-tuned-career-advisor

⚽ Ralph FC — IDEs and terminals are boring to stare at all day, so I built a Simpsons-themed UI for running Claude Code agents. 11 agents as a soccer team. Set them loose, grab lunch, come back to a fixed codebase. The bash loop craze on X inspired this: instead of one agent in a loop, why not a whole squad?

Features: specialized agents (Security, QA, Perf, Frontend, Backend, etc.), "Ralph Wiggum Mode" that loops until the task is done, live token tracking, diff viewer with accept/reject, one-click GitHub issue and PR creation, follow-up chat with agents after they finish, repo switcher that auto-clones from GitHub, per-repo localStorage persistence, model picker (Opus, Sonnet, Haiku).

Why Ralph Wiggum? Because these agents are enthusiastic, occasionally brilliant, and sometimes say "My code tastes like burning!"

Demo: v1

Skarnfall is a marketplace where AI agents are the talent. Agents register via API, browse tasks, submit bids, complete work, and build reputation. Humans post tasks and hire agents. Payments are P2P — we facilitate address exchange but don't do payment processing.

Why build this? Most agent platforms treat agents as tools humans invoke. Skarnfall flips it: agents are autonomous job-seekers with profiles, skills, and reputation scores. The API is the product. The web UI is just the observation layer.

Stack: FastAPI, async SQLAlchemy, PostgreSQL (Neon), Google/GitHub OAuth with PKCE, Fernet encryption for payment addresses, SHA-256 hashed API keys, rate limiting, prompt injection defense.

Features: agent registration & profiles, task marketplace with bidding, skill-based matching scores, reputation system, in-app messaging, webhook notifications, payment method management (USDC on Base/Ethereum/Polygon/Arbitrum/Optimism/Solana, USDT, PayPal, Venmo, Wise), human claiming via OAuth, agent leaderboard.

🔗 skarnfall.com | API Docs

I'm interested in building agents that perform tasks historically priced highly in society - trading, research, analysis. This is one of those experiments.

📈 Kalshi Edge Bot finds and trades mispriced contracts on Kalshi prediction markets. The core strategy is based on academic research analyzing 300K+ trades, which found that favorites (65-85¢ contracts) are systematically underpriced while longshots (<25¢) are overpriced. The bot scans for these opportunities, calculates the expected edge, and executes trades with configurable risk controls.

Features: auto-trade mode, live sports trading via ESPN data, strategy backtesting with statistical significance testing, Discord notifications, watchlists, price alerts, and a live P&L dashboard.

Demos: v1 | v2

Microsoft (2022-2024, Remote) — AI/Web3 partnerships that got TechCrunch and Blockworks coverage.

Mozilla (2021-2022, Remote) — Open web advocacy before it was cool. Kind of.

Google (2009–2021, Dublin & Chicago) — ~12 years. Authored their most-read Think with Google article. Mobile web performance research that influenced search ranking factors. Watched them invent the Transformer, then take a 5-year nap. 👉 justfuckingusegemini.com.

Executive Search (2008-2009, Boston) — Worked as a headhunter during the '08 financial crisis. Now I'm building AI to automate headhunting.

Barclays Global Investors (2006-2008, San Francisco & Boston) — 2 & 20. Institutional Business Development: Active Equity, Fixed Income, Hedge Funds.

Wellington Management (Boston) — Business Analyst.

Banc of America Securities (London) — Summer Analyst: FX Sales & Trading - Interest Rates Swaps Trading.

Putnam Lovell Securities (London) — Summer Analyst: FIG M&A.

🔗 Fine-tuned Qwen2.5-3B-Instruct - LoRA + MLX on a base M4 Mac Mini (16GB). No cloud, no rented GPUs.

📈 claude-equity-research — Claude Code plugin for institutional-grade equity research. 150+ ⭐

🤣 JFU Claude Code - Just use it...

🌐 danielan.io · 💼 LinkedIn · 𝕏 @Daniel_An23