KENZY.Ai

Still building for “user experience” in 2026? Cute. AI agents handle the busy work now. Your real customer is becoming software. If your product’s value disappears when a human isn’t present, that’s not a future-proof business.

Biggest problem with the current incarnation of Kenzy is picking up voice clear enough and getting rid of background noise.

Considering refactoring Kenzy.Ai into a #smartspeaker in #python. Yes, I know about the HA Voice PE… and I just don’t care. I need something I can control completely from the ground up. Wanna help? @SmartHomeSolver @EverySmartHome @geerlingguy

Likely streaming audio to/from a beefier machine rather than doing everything local at the edge (those GPUs are expensive). Also makes it more portable for those interested in cloud hosted solutions like OpenAi.

Deeply familiar with OpenAI Whisper for local STT. Probably some variety of deterministic intent evaluation and LLM backup. For TTS the most realistic local voice I’ve found is SpeechT5, but Piper works too and there are others that I’m less familiar with. Open to ideas.

Hardware Problems: I need a far field mic, possibly with beam forming for capture. I need a reasonably powerful speaker. And I need really, really good AEC between them. I also will likely enable LLM, and I want a fully local option (probably mistral/ollama).

@holaivory I’ve built tons, and I’d argue that unless you’re going top shelf for everything the offerings in the market today for mid-level prepackaged hardware are likely just as good and cheaper. That being said, experience is worth its weight in gold if it’s your first time.

@ImLunaHey Yes. Oracle, SQL Server, MySQL, MariaDB, Postgres, Redshift, SQLlite, DB2 (zOS and LUW), and a bunch of NoSQL. Some may say I work in the data space. Jack of all databases… master of a few.

what database do you use? 👇

@ryanels COBOL… haha

What language am I coding in 🤔

I cannot even remember the last time I installed windows from scratch… so many bad memories.

@matifanger All these fruity bananas on here… lol Dude, it’s pizza AND nachos. For the health-minded individuals, my mental acuity is not normally diminished by snacks when sufficiently challenged with a hard problem. In fact it’s hard to sleep at all as my brain works the solution.

Best coding snack: nachos or pizza?

@engineers_feed Empty both. start both. Stop 7 when 4 ends. Flip 7 and 4 so that it drains the 3 and stop 4 leaving 1 in 4 and zero in 7. Flip both and stop 7 when 1 is out of 4. Leaving 6 in 7 and zero in 4. Flip 4 and let 7 drain until 2 remain. Flip 4 again until 2 remain. Now 7+2=9

Logic puzzle.

New “Mute” skill added in prep for the fallbacks so you can stop runaway conversations if they occur… gotta be ready for whatever this LLM tosses at me. This will be included in the upcoming 2.1.4 release.

@OngDevLab Test (code changes) in dev and test. Monitor in prod. Test (scenarios) in prod (via red/green).

Definitely in need of some cable management. 😮

It took a while, but it’s working… and the GPU is noticeably faster than the mini PCs. Next task is building a fallback to use this… then maybe see about hardware upgrades.

I pulled out my z420 today because it has a NVIDIA GPU and 96GB of RAM in it for some more LLM testing for Kenzy… fun stuff.

Just putting this here for safe keeping… ;-) youtube.com/live/jJKbYj8mI…

@danielcranney Inner peace… or really good over the ear headphones.

People who code in silence, without music or podcasts, what’s your secret?

@OngDevLab Di.fm. You’re welcome.