Signal Notes

The old guard is out.

The headlines make it sound like custom AI needs a bank of supercomputers and a team of PhDs. What if it doesn’t?

Build Log, with Nick Creighton.

A quiet story of shipping. This week, we move past the hype to the real workbench. The goalposts have moved. We’re talking about fine-tuning that’s faster, cheaper, and shockingly accessible—practical for the rest of us, running in the background of everything we build.

Full breakdown: [Link to blog post]

See how it fits together. Listen to Build Log.

Build Log. Nick Creighton.

A quiet rebellion against the cloud. The real cost of AI isn't in the API docs—it’s in the monthly bill. Nick just pulled his AI workflow in-house, deploying a private agent for his entire content network. The price tag? Under fifty bucks.

This is about taking back control. It’s the hum of a local server, not the silent drain of a metered service. A breakdown of the hardware, the models, and the math they don’t want you to see. The vibe is autonomy.

For the full cost analysis, see the companion post.

Listen to the quiet hum of your own machine.

Build Log with Nick Creighton.

A demo that feels flawless can lie. Deploying a RAG system taught me that the hard way. Real users encountered confident, made-up answers.

This episode is a wake-up call: trading gut checks for hard metrics. We move from art to science. From holding your breath at launch to trusting what you've built. It’s about finding the three numbers that tell you the truth before your users do.

Stop guessing. Start measuring.

Dive deeper in the companion blog post. Listen now.

The cloud bill that broke the camel's back.

Host Nick Creighton turns away from the API roulette wheel and into the quiet hum of local hardware. This is the off-grid manifesto for practical AI: the tangible clunk of a server in a closet, the silent blink of an LED on a SBC, and the stark reality of cost sheets compared to latency graphs. It's about reclaiming inference from the distant data center, finding the raw edge in your own rack, and the machines that make it possible without vaporizing your budget.

A guide to the gear that actually works when real revenue is on the line.

Dive deeper with the companion blog post: [link]

Listen to the quiet revolution.

A high-performance AI, running entirely offline on a five-year-old laptop. This isn't a dream of the future; it’s an operational reality. Nick breaks down the quiet revolution of quantization—the technique that shrinks massive language models by 75% with almost zero loss in performance. It’s about moving from a recurring cloud utility bill to a deployable asset you truly own. The vibe is liberation: from pricing, from privacy concerns, from the cloud. It’s the sound of capability being unlocked.

Dive into the technical details on the companion blog post.

Listen to Quantizing Llms For Local Ai 2024.

This week, Nick shipped something that works while he sleeps. It’s an automated research agent—not a chatbot, but a digital employee. It quietly sifts through dozens of sources for a single data point, costing less than coffee per week. This is about leveraging the new reasoning of LLMs to build a true assembly line for knowledge. A small, reliable system that scales, turning the tedium of manual research into a quiet, automated process.

The full build log is documented [here].

Listen for the quiet hum of a system that just works.

A seismic shift in the open-source model landscape. Llama 3 8B is the new darling, but the data from thousands of real tasks tells a different, quieter story.

This episode is a rollback. A correction to the herd mentality. Host Nick Creighton shares a costly personal misstep—weeks and budgets lost chasing the obvious new thing—to argue for the nuanced, practical choice.

It’s not about abstract benchmarks. It’s about the right tool for your stack, your costs, your deployment clock. The vibes are glitchy, pragmatic, and pulled straight from the production logs.

Dive into the full analysis on our companion blog.

Listen to find your true north.

No API calls. No subscriptions. No data leaving your machine. This is the quiet, powerful alternative to cloud AI coding.

I built a local AI assistant with Ollama. It’s private, free, and runs entirely offline. My code never touches another server. This changes everything—especially for sensitive client work where compliance and privacy are paramount.

Forget the monthly fees. This is about total control and quiet confidence, right from your own terminal.

Read more details on the build in the companion blog post.

Listen to the full workflow.

Nick realized the real bottleneck in AI isn't the models, but the tedious, manual work of moving *between* them. He was spending hours copy-pasting data between tools every Monday.

This episode is about moving from playing single notes to conducting entire symphonies of AI. We explore the minimalist, practical architecture for orchestrating workflows that *actually works* in production.

Ditch the busywork. Start conducting.

Read the companion post and listen to the full episode.

This isn't a hypothetical. We saw enterprise RAG hallucinate daily. A raw, unchecked answer could spark a lawsuit.

We spent three months across 13 live sites chasing the ghost in the machine. Here’s the blueprint we built: an AI fact-check system that slashes errors by 85%, no custom code required. We’ll walk through the exact architecture, from the subtle electronic pulse of a trigger to the final, verified output.

It’s about trust, verified. Find the full technical deep dive [here].

Listen for the signal, not the noise.