Signal Notes

Nick Creighton

Dispatches from a 13-site AI empire — what actually works in production, what fails, and what nobody tells you about building with AI.

  1. 22 jun

    Local Rag With Ollama Tutorial

    The hype around local AI often crashes into the reality of ten-gigabyte downloads and cooling fans spinning like jet engines. It’s a promise of privacy that feels violently out of reach. In this episode, we bypass the academic noise and hardware meltdowns. We are deploying an offline Retrieval-Augmented Generation system that breathes life into your documents without sending a single byte to the cloud. No API keys. No monthly invoices. Just a clean, efficient stack running on a standard laptop. It is the sound of productivity, reclaimed from the chaos of complex tutorials. Get the written walkthrough at [Link] Listen to build your local archive.

    13 min

  2. 22 jun

    Ai Agent Frameworks Eval 2024

    The carnival rides are shutting down. The demos are over. Building AI agents in 2024 on nothing more than a loop and a prayer isn't just risky—it's a structural failure waiting to happen. We’ve seen the infinite loops. We’ve burned the API credits. We’ve watched hallucinations compound into total system collapse. This is the shift from hype to reliability. We strip away the magic to look at the frameworks that actually keep production systems running when things get weird. It’s time to stop hoping your prompt works and start building a house that stands. Read the companion blog post. Press play and let’s ship.

    15 min

  3. 15 jun

    Fine-Tune Llama 3 706B Model Locally

    The hum of a server. The click of a terminal. Forget the API; this is about running the colossal Llama 3 model yourself. A 706B parameter AI, in-house, offline. Host Nick Creighton maps the frontier of true data privacy, where your internal documents and codebases never leave your control. This is the blueprint for building a private AI data center, covering the hardware reality check and the parallel power of fine-tuning and RAG. Ready to ship it? Listen here. Read the companion post: [Link to blog post]

    17 min

  4. 15 jun

    Fine Tune Llama 3 For Sql Query Generation Tutorial

    You trust the AI to write your SQL. It's going to cost you. This week, we fine-tune Llama 3 in an afternoon for under fifty dollars, turning a brittle liability into a precise, private tool. It’s about the gap between a convincing answer and a correct one. The quiet panic when a hallucinated column sends your dashboard off a cliff. We’re moving past generic chatbots and into something that understands your schema, your logic, your stakes. Build the detail-oriented, credible alternative. Full walkthrough in the companion post. The next query it writes could save your morning. Listen.

    15 min

  5. 8 jun

    Llm Evaluation Metrics Explained 2024

    Build Log, with Nick Creighton. This week, the models went quiet. The outputs, once reliable, turned bland and hollow. When your systems falter and hope is your only strategy, it’s time to move past the demo. Nick recounts the death of the "vibe check"—that quick, gut-feeling review that fails when you’re not looking. He spent the last three months building a real validation pipeline, shifting from fragile prompts to a system that actually earns its keep. This is about fighting the silent decay of AI performance, about replacing theory with a foundation that holds while you sleep. For more detail on the validation build, find the companion post [link]. Listen to the full episode.

    9 min

  6. 8 jun

    Openai Api Vs Local Llama 3 Cost 2024

    Signal Notes. March 25th, 2024. A cold number on the dashboard at dawn. The hum of a server, the quiet click of a key. The cost of intelligence is plummeting, a 92% freefall in 14 months. The gap between cloud and local inference has narrowed to a sliver, a decimal point on a billing report. The raw data from Nick’s production run: $347.22 for the API, $412.00 for the rented hardware, plus the hidden tax of maintenance scripts and library conflicts. It’s a story told in tokens and receipts, not theory. A vibe of pragmatic calculation. The quiet awe of a shifting landscape. Read the numbers: [companion blog post link] Listen to the math.

    25 min

  7. 5 jun

    Ai Agent Frameworks Vs Traditional Automation 2024

    The old map is obsolete. It's being replaced by a compass. Traditional automation is a brittle, minimum-wage workforce. AI agents are something else entirely—navigators that work off-road. Nick put both systems to the test over three months across thirteen sites. The results weren't close. It’s a fundamental architectural shift happening right now in production, moving from rigid step-by-step processes to adaptive, goal-oriented execution. This isn't a future prediction. It's a present-tense reality measured in hours saved and systems that don't break. Dive into the data and the details in our companion post. Grab your headphones. Let's build.

    9 min

  8. 5 jun

    Rag Vs Fine-Tuning For Document Qa 2024

    Build Log. I’m Nick. If you’re using a fine-tuned model for document Q&A, you’re likely burning cash for worse results. This is the critical build-vs-buy decision for your AI’s brain, and it’s a weekly invoice that decides if your project lives or dies. GPU costs are falling, but fine-tuning API prices haven’t. The real killer? Knowledge cutoffs. A perfectly formatted, completely wrong answer from a model trained on last year’s docs. RAG solves this inherently. New docs hit the vector store, and seconds later, your AI knows. No retraining. No extra cost. A three-month production test. The winner wasn’t close. Read the full breakdown on the blog. Listen to the episode.

    9 min
Ver todo (34)

Acerca de

Dispatches from a 13-site AI empire — what actually works in production, what fails, and what nobody tells you about building with AI.

Información

  • Creador
    Nick Creighton
  • Años de actividad
    2 k
  • Episodios
    34
  • Clasificación
    Apto
  • Copyright
    © (c) 2026 Aevorn LLC
  • Mostrar sitio web
    Signal Notes