📓 Creative Notebook for Nov 24 - Taxes, AI, Personas and Bluesky

This week was a whirlwind of AI, new tools, and even a dive into tax preparation using a locally ran large language model. Here’s what I’ve been working on, learning, and building over the last week.

Happy Data AI Tools

I kicked off the week by publishing happydata-ai, a small yet powerful set of utilities I rely on for nearly every AI-driven project. It’s designed for OpenAI-compliant endpoints and Ollama, with features like forced JSON output using Zod Schemas, variable prompts, PII redaction, and more.

Should you use it? Honestly, I’m not sure—but I do, and it makes my life easier. Check it out if you’re diving into AI integrations.

Got on BlueSky

I finally decided to leave the madness of the platform formerly known as Twitter for BlueSky, and let me tell you—it’s a breath of fresh air. No echo chambers, just meaningful connections. If you’re there, let’s connect: brandons.app.

Prepping for Taxes with AI

Yes, you read that right: I’m tackling tax prep with AI. I started building a tool using Bun and Ollama to run Llama 3.2-instruct locally on my Mac, ensuring nothing private goes over to OpenAI.

Here’s the process so far:

1. Search Gmail for receipts, invoices, and POs.
2. Extract attachments and HTML emails, then convert everything to Markdown.
3. Pass the Markdown and my Zod Schema to Llama 3.2 to generate JSON payloads breaking down each item.

So far, it’s working! Next week, I plan to convert the JSON data into a SQLite database, where I’ll query and manage it using a tool I built a few weeks back.

InfoCap AI Seminar

This week, I flew to Atlanta to attend InfoCap’s AGENTIC AI + Decision Intelligence seminar. It was a small but impactful event, even if a bit of a soft ad for InfoCap.

I walked away with some fresh ideas, particularly around Directed Acyclic Graphs (DAGs) and the OODA loop (Observe, Orient, Decide, Act). These concepts are already influencing how I approach problem-solving and decision-making in my projects.

Persona Generation Adventure

About six months ago, a major client asked me to generate customer personas based on their data. After a five-month wait, I finally received a dataset… but couldn’t confirm who the data represented. I ran the process, generated personas, and prepped for a leadership pitch.

Then, my boss asked, “Can you confirm where this data came from?” (Pro tip: Always do this first!) Turns out, the dataset contained 70,000 non-profit members, most of whom were not customers.

I reworked the pitch, showcasing the methodology, but made it clear we needed real customer data. Shortly after, I gained access to 400,000 actual customer records and delivered a polished set of personas that resonated with the leadership team. Lesson learned!

Looking Ahead

From launching AI tools to exploring new platforms and applying decision intelligence frameworks, this week has been equal parts challenging and inspiring. Next week, I’ll be diving deeper into the tax tool, refining personas, and exploring even more creative AI applications.

Let me know what you think of these projects—or better yet, share what you’re working on. Let's build something amazing!