Joe Freeman
@fka-freemansoft.bsky.social
On YouTube: https://www.youtube.com/@FreemanSoft On LinkedIn: https://www.linkedin.com/in/1freeman/
youtube.com/shorts/ksBPk... An E-Ink mood display - Mood Badger 2350 - for people who can't visually express emotions or people who work with others who cannot read other people's emotions.
Pimoroni Badger 2350 Badgeware E Ink Mood Display
YouTube video by Joe Freeman: Software Craft, Org Stuff, Tech Stuff
youtube.com
Move beyond just spewing business functionality across your service or application. Create a common understanding of how your business rules can be declared, how they are used, and how you can apply discipline to their organization. youtu.be/sZMgGfyFOAw
Defining rules, rulesets, and happy and unhappy path behavior around their execution and consumption
YouTube video by Joe Freeman: Software Craft, Org Stuff, Tech Stuff
youtu.be
Effective Dates and Transaction dates give operational systems a Bi-Temporal nature. - Transaction or Activity Date - Effective or Valid Dates youtu.be/2QAkTwHTx9g #Lakehouse #Bi-Temporal
Bi-temporal - Effective Date and Transaction Date in Transactions
YouTube video by Joe Freeman: Software Craft, Org Stuff, Tech Stuff
youtu.be
Data Lakes are often three-time-axis, or Tri-Temporal, in nature when storing business operations and transactions. - When did it commit - When was it active or in force - When was it visible in the lake #Lakehouse #Transactions #DataLake youtu.be/gTliu0T8eps
The Tri-Temporal Nature of Data Lakes when holding Bi-Temporal Operational Data
YouTube video by Joe Freeman: Software Craft, Org Stuff, Tech Stuff
youtu.be
🚨 LAST CHANCE TODAY! The discounted hotel room block rate at the Baymont by Wyndham for #BSidesDE2026 ends tonight! 🏨🧳 Group Code: GS111226SEC Need a roommate? Use #BSidesDERoomShare! 🌐 Book now: www.bsidesdelaware.com/venue/ #InfoSec
Venue and Hotel Information - Security BSides Delaware
BSides Delaware is back at the Fintech Building at the University of Delaware in 2026!
bsidesdelaware.com
Save the date! BSides NoVA is back for our 10th year! Join us October 30-31 as we celebrate a decade of community, collaboration, and cybersecurity. We're just getting started, so stay tuned for this year's theme, CFP, Happy Hours, and more announcements in the coming weeks.
My blog whispered 'genius'. Turning the content into a video shouted 'chaotic dreck'
In Ollama, New conversation cache behavior depends on the model's memory and the runner. Seven attention-only models reused the prompt. Four on llama-server always re-evaluated a fixed 1024-token. MLX runners paid one full cold prefill and then nothing. joe.blog.freemansoft.com/2026/09/on-o...
On Ollama, a model's memory type and its runner decide how much of the cached prompt a new conversation reuses - 8
In freemansoft/Flutter-AdaptiveCards a demonstration Dart chat server asks a local Ollama model for an answer as Adaptive Card JSO...
joe.blog.freemansoft.com
Eight local models on Ollama answered the same 25 questions twice, once with an empty context window and once carrying roughly 48,500 tokens of history: Three lost a quarter to a third of their coverage. joe.blog.freemansoft.com/2026/09/a-fu...
A full context breaks three local models, each in a different way - 7
In freemansoft/Flutter-AdaptiveCards a demonstration Dart chat server hands a question to a local Ollama model. It asks for the an...
joe.blog.freemansoft.com
If you send a local Ollama more chat history than its context window holds, Ollama doesn't trim history. It throws the whole message away, returns a normal-looking answer, and reports no error or warning. I observed this on 14 models and 2 Ollama versions. joe.blog.freemansoft.com/2026/09/olla...
Ollama silently drops a history message larger than its context window
In freemansoft/Flutter-AdaptiveCards a demonstration Dart chat server hands a question to a local Ollama model. It asks for the an...
joe.blog.freemansoft.com
To measure model behavior, a wave of probes sent a fixed set of questions to Ollama, one model at a time. Several results looked like bad model behavior when the root cause was test config, the machine, the Ollama runtime, the harness or, the test probe. joe.blog.freemansoft.com/2026/09/eigh...
Eight measurement rules from a local-model benchmark on Ollama
In freemansoft/Flutter-AdaptiveCards a demonstration Flutter client sends questions to a Dart chat server. The server passes each one ...
joe.blog.freemansoft.com
Ollama's tool channel beats prose for card JSON on every model that calls it. A structured response: the JSON comes back in the model's message body as JSON text inside the message content (as prose). Ollama also offers a second route, the tool channel. joe.blog.freemansoft.com/2026/09/olla...
Ollama's tool channel beats prose for card JSON on every model that calls it
In freemansoft/Flutter-AdaptiveCards a demonstration Dart chat server hands a question to a local Ollama model. It asks for the an...
joe.blog.freemansoft.com
joe.blog.freemansoft.com/2026/09/i-ac... A Fable: An SDK upgrade I've done repeatedly has a long tail. The tail includes triaging golden-image failures and managing the images. It includes pushing CI-generated images back into the repo. Today, the agent did the whole workflow unprompted.
I accidentally used Fable to do a system upgrade and it was Fableous
A Flutter SDK upgrade I've done several times has a long manual tail that I've never been able to hand off. The tail includes deciding whic...
joe.blog.freemansoft.com
We ran into 4 types of failures when building an SDUI demo that used an LLM as an AdaptiveCard JSON generator. Raw JSON displayed as text. Truncated Cards truncated. Prose appended after an otherwise valid card. A conversation that drifts back to prose. joe.blog.freemansoft.com/2026/09/we-t...
We tried 14 levers to get reliable SDUI card JSON from a local LLM model
In freemansoft/Flutter-AdaptiveCards a demonstration Dart chat server hands a question to a local Ollama model, asks for the answe...
joe.blog.freemansoft.com
Adaptive card generation turned out to be an interesting test problem for local models. The results carry beyond this demo to any workload that asks a local model for constrained, schema-shaped JSON. The demo became an instrumented test bed. joe.blog.freemansoft.com/2026/09/an-s...
An SDUI demo that turned into a local-model benchmark
freemansoft/Flutter-AdaptiveCards is a Flutter renderer for Adaptive Cards. The project includes a demonstration Flutter chat client...
joe.blog.freemansoft.com
Sometimes I hate what coding LLMs represent. Other times I love the fact that they let me push projects to completion with more features, more tests, and more documentation. joe.blog.freemansoft.com/2026/08/soft...
Soft entity resolution using ElasticSearch and data analysis: DOT chameleon-carrier detection as a PoC
A few years ago I tried to build a probabilistic/soft entity resolution system a few years ago using Elasticsearch with its ability to match...
joe.blog.freemansoft.com
Adaptive Cards is one of the methods bots and team spaces use for delivering cross platform Server Specified UI. I demonstrate how this could work using a Flutter adaptive card library, a lightweight server and Ollama. youtu.be/ek45Q5rcYQQ?...
Flutter Adaptive Cards Ollama Chat demonstration - Agent generate UI.
YouTube video by Joe Freeman: Software Craft, Org Stuff, Tech Stuff
youtu.be
The documentation is the code, and the LLM is the compiler. Specs used to be written once and ignored. With LLMs, they become shared state - drafted, revised, and reconciled by humans and agents every time the system changes. youtube.com/shorts/Ru8qz... joe.blog.freemansoft.com/2026/06/my-b...
Specs Are the Code when it Comes to Agentic Coding
YouTube video by Joe Freeman: Software Craft, Org Stuff, Tech Stuff
youtube.com
The LILYGO T-Dongle-S3 is self-contained, including a microcontroller, Wi-Fi, Bluetooth, a Stemma/I2C, an SD Card slot, and an LCD display. It can be used as a standalone device, a computer-attached peripheral, or a standalone cloud endpoint. #circuitpython #adafruit #lilygo youtu.be/rCJWZwMG2nQ
Implementing standalone or tethered IoT appliances with the LILYGO T-Dongle-S3 and CircuitPython
YouTube video by Joe Freeman: Software Craft, Org Stuff, Tech Stuff
youtu.be
#Adafruit has a #CircuitPython package for the #LILLYGO T-Dongle-S3 that uses the LCD display as its REPL output. You can see this by enabling the LCD backlight in your Python. You can override this and use the LCD for your own program with a couple of lines of Python. youtube.com/shorts/K2Tz9...
Writing Python code for the LILLYGO T-Dongle S3 LCD Display with Adafruit CircuitPython
YouTube video by Joe Freeman: Software Craft, Org Stuff, Tech Stuff
youtube.com
Watching the NVIDIA GTC DC pre-game talk. The entire discussion revolves around building data centers, their power requirements, and the country's need to invest in infrastructure for AI factories. Massive power consumption isn't a problem. It's a feature. #GTC #NVIDIAGTC #GTCDC
Attended Sean Metcalf's BSidesNoVa 2025 talk "10 ways to Improve Entra ID Security Quickly" pyrotek.io/BSNoVa2025. My main takeaway was that some of the default settings are bad, that Microsoft has been iterating, and that Securing #EntraID (FKA Azure AD) requires serious expertise. adsecurity.org
pyrotek.io
A big shout-out to Pavan Reddy, who gave a great lab session at the BSides NoVa event in Northern Virginia bsidesnova-2025.sessionize.com/session/992557 The entire session ran in Google Colab, launched with a single click, running nothing on the local machine. The only requirement was a Gmail ID.
How to Break an AI: Adversarial Attacks, Jailbreaks, and Defenses That Actually Work
AI systems don’t break like software, they fail in silence, misclassify with confidence, and hallucinate under pressure. This 4-hour hands-on workshop exposes the core vulnerabilities of modern AI, fr...
bsidesnova-2025.sessionize.com
The case against generative AI as a business and the questionable revenue hype machine. This article, which I wish I had written, describes the tangled web of hype, power, and corporate deals driving today's Generative AI. www.wheresyoured.at/the-case-aga...
The Case Against Generative AI
Soundtrack: Queens of the Stone Age - First It Giveth Before we go any further: This is, for the third time this year, the longest newsletter I've ever written, weighing in somewhere around 18,500 wo...
wheresyoured.at
DevFestDC is back. October 3, 2025, in Arlington. www.devfestdc.org #DevFest #GoogleDeveloper #AI
DevFest DC 25
DevFestDC is an annual, community-led software engineering conference designed to educate software developers, architects, entrepreneurs, quality-focused individuals, and tech leaders.
devfestdc.org
BSides NoVa: BSides Northern Virginia is coming up October 10-11 2025. Can't get to DEFCON? Need another security fix? Attend a local BSides #BSides #DEFCON #Hacker www.bsidesnova.org
Admissions for Hacker University 2025 are open – Apply Now!
bsidesnova.org
Data science and AI projects are typically run on rented CPU/GPU resources or in self-managed data centers. By using containerization, we can separate our code and configuration from the underlying hardware youtu.be/pFH_Yk-cGec?...
Visiting the topology when running remote containerized data science workloads and AI Workbench
YouTube video by Joe Freeman: Software Craft, Org Stuff, Tech Stuff
youtu.be
The asustor AS5402T is a NAS appliance that supports two (2) 3.5" drives and four (4) NVMe drives. I wanted to run some Docker and Home Automation on my AS5402T NAS. Upgrading from 4GB to 32GB was a simple, toolless operation. joe.blog.freemansoft.com/2025/09/upgr...
joe.blog.freemansoft.com
