|
17298
|
GardenBuddy by @Vedant_Daga | Stardance
|
Vedant_Daga
|
0
|
3.0
|
0
|
[]
|
[{"post_id": 8784, "ts": "2026 [{"post_id": 8784, "ts": "2026-06-14T16:23:29Z", "hours": 3.0, "text": "gardenbuddy devlog #1 i kept killing plants. not from neglect, just from guessing wrong. water too much, roots rot. too little, leaves curl. i never actually knew what was happening underground. so gardenbuddy is a raspberry pi that watches soil, temp, humidity, and light, and tells you what to do about it. most plant monitors just give you numbers. 48% soil, 72°F. okay, is that good? i wanted it to answer that directly. not “soil is at 48%” but “your plant is stable, humidity is the main stressor, here’s what to do.” how it works two models. an LSTM looks at the last 30 readings (5 min) and classifies health as thriving, stable, stressed, or critical. it trains on your own sensor data, so it learns your setup instead of using thresholds someone else picked. a local LLM (ollama, llama3.2:3b) takes that output and writes two plain sentences you can act on. runs every 30 seconds in a background thread so it doesn’t block polling. all local. no cloud, no api keys, no internet. what’s running live readings every 10 seconds, rolling history charts, a 0-100 health score, the LSTM classification, ollama advice, an RGB LED that goes green/amber/red, influxdb for history, and alerts for low soil, heat, and fungal risk. the cleanup once the AI side worked, sensor_reader.py was 454 lines. about 200 were fallback paths and stubs i’d written just in case the model didn’t load or ollama timed out. useful while building, unreadable once things were stable. deleted them: 13 files, 219 lines gone, zero new features. that’s kind of when it started feeling real. hardware pi 4b, DHT22 for temp/humidity, capacitive soil sensor through an ADS1115 ADC over I2C, a photoresistor for light, and the RGB LED. soil calibration is two numbers you measure once: dry air and fully submerged. everything else comes from those. next the foundation works. what’s missing is a model that’s seen more than a few days of data, and a real stress test of the alerts. plan is to underwater a plant on purpose and watch the score drop. apparently it’s doing fine."}]...
|
Title: GardenBuddy by @Vedant_Daga | Stardance
De Title: GardenBuddy by @Vedant_Daga | Stardance
Devlogs:
gardenbuddy devlog #1 i kept killing plants. not from neglect, just from guessing wrong. water too much, roots rot. too little, leaves curl. i never actually knew what was happening underground. so gardenbuddy is a raspberry pi that watches soil, temp, humidity, and light, and tells you what to do about it. most plant monitors just give you numbers. 48% soil, 72°F. okay, is that good? i wanted it to answer that directly. not “soil is at 48%” but “your plant is stable, humidity is the main stressor, here’s what to do.” how it works two models. an LSTM looks at the last 30 readings (5 min) and classifies health as thriving, stable, stressed, or critical. it trains on your own sensor data, so it learns your setup instead of using thresholds someone else picked. a local LLM (ollama, llama3.2:3b) takes that output and writes two plain sentences you can act on. runs every 30 seconds in a background thread so it doesn’t block polling. all local. no cloud, no api keys, no internet. what’s running live readings every 10 seconds, rolling history charts, a 0-100 health score, the LSTM classification, ollama advice, an RGB LED that goes green/amber/red, influxdb for history, and alerts for low soil, heat, and fungal risk. the cleanup once the AI side worked, sensor_reader.py was 454 lines. about 200 were fallback paths and stubs i’d written just in case the model didn’t load or ollama timed out. useful while building, unreadable once things were stable. deleted them: 13 files, 219 lines gone, zero new features. that’s kind of when it started feeling real. hardware pi 4b, DHT22 for temp/humidity, capacitive soil sensor through an ADS1115 ADC over I2C, a photoresistor for light, and the RGB LED. soil calibration is two numbers you measure once: dry air and fully submerged. everything else comes from those. next the foundation works. what’s missing is a model that’s seen more than a few days of data, and a real stress test of the alerts. plan is to underwater a plant on purpose and watch the score drop. apparently it’s doing fine....
|
1792946958.9253092
|
1793810958.9253092
|
|
38220
|
Bamboo by @panda | Stardance
|
panda
|
0
|
0.0
|
0
|
[]
|
[{"post_id": 32927, "ts": "202 [{"post_id": 32927, "ts": "2026-07-22T12:23:42Z", "hours": 0.0, "text": "Hey - I just had an idea for a brand new project The Problem Ok, i bet you’ve used AI before - maybe even an AI agent ? I definetely have - I use my opencode agent almost every day for something or the other. Problem is, AI isn’t reliable - if it doesn’t know what its doing, it’ll confidently tell you to blow yourself up sky high as a method of flight and spawn slop from the deepest pits of hell. It lies so confidently that you believe it, cause you might be way out of your depth and just wanna get something working. The A solution to your problem would be to just learn the things yourself. But be honest - when has that ever worked ? While its definetely worthwile to learn somethin new, its also time consuming and you don’t always have that time. And sometimes you just have problems where the situation is so specific that you either learn an entire new programming langiage just for that one obscure thing or use an unreliable AI agent. By now your probably asking yourself where im going with this. Thats where my idea comes in. Instead of having just one AI agent spawn some random bullshit, use another one - use a second agent with a second provider that double checks what the first one does - instead of just one agent, have multiple so they can split up tasks to make efficiency higher, have them cooperate on tasks and cross-check what the other ones are doing and make sure you don’t get AI slop that blows up in your face ! This is the vision. All i did right now was ask Opencode if this was a reasonable idea, and asked it to make a “Framework” or “Scaffolding” with just an outline I can build upon. See ya soon - and if you took the time to read this, i greatly appreciate you - let me know what you think or if you have any ideas for me - i would love to hear from you !"}]...
|
Title: Bamboo by @panda | Stardance
Devlogs:
Hey Title: Bamboo by @panda | Stardance
Devlogs:
Hey - I just had an idea for a brand new project The Problem Ok, i bet you’ve used AI before - maybe even an AI agent ? I definetely have - I use my opencode agent almost every day for something or the other. Problem is, AI isn’t reliable - if it doesn’t know what its doing, it’ll confidently tell you to blow yourself up sky high as a method of flight and spawn slop from the deepest pits of hell. It lies so confidently that you believe it, cause you might be way out of your depth and just wanna get something working. The A solution to your problem would be to just learn the things yourself. But be honest - when has that ever worked ? While its definetely worthwile to learn somethin new, its also time consuming and you don’t always have that time. And sometimes you just have problems where the situation is so specific that you either learn an entire new programming langiage just for that one obscure thing or use an unreliable AI agent. By now your probably asking yourself where im going with this. Thats where my idea comes in. Instead of having just one AI agent spawn some random bullshit, use another one - use a second agent with a second provider that double checks what the first one does - instead of just one agent, have multiple so they can split up tasks to make efficiency higher, have them cooperate on tasks and cross-check what the other ones are doing and make sure you don’t get AI slop that blows up in your face ! This is the vision. All i did right now was ask Opencode if this was a reasonable idea, and asked it to make a “Framework” or “Scaffolding” with just an outline I can build upon. See ya soon - and if you took the time to read this, i greatly appreciate you - let me know what you think or if you have any ideas for me - i would love to hear from you !...
|
1792946937.460161
|
1793810937.460161
|