Where simple problems meet gloriously disproportionate solutions.
A growing collection of projects where the solution is unapologetically disproportionate to the problem. Every project in this series exists because "but I could automate that with AI" is a lifestyle, not a suggestion. Efficiency? Optional. Style points? Mandatory.
A ₹500 voltage relay would work. Instead: a Raspberry Pi Pico W streaming JSON telemetry over WebSockets to a FastAPI server hosting an Ollama LLM agent that maintains belief states, runs an OODA loop, and calls Python tools to flip a relay. Also reflects on past decisions and grades its own performance. The relay has no feelings about this.
A spray bottle would keep plants alive. Instead: a Pico W with an SHT41 precision sensor, 4-channel relay board, Python backend, React + Framer Motion dashboard with an animated terrarium visualizer, and a Gemini AI that performs compatibility analysis before you're allowed to add a new plant. The ferns did not ask for this level of vetting.
A ₹999 IKEA sensor shows three colors: green, yellow, red. Unacceptable. So: hack it open, wire it to an ESP32 C6, flash custom Tasmota, collect 11 months of atmospheric telemetry, run it through a 12-stage Python pipeline with Hatch humidity correction, Lomb-Scargle periodograms, Gaussian dispersion modeling, and Facebook Prophet forecasts. Also 3D-print a custom HEPA purifier. Obviously.
The rate limit banner appears. Unacceptable. So: a Manifest V3 Chrome extension that injects into the Main World to wiretap undocumented /usage API traffic, silently extracts your Org UUID, auto-rotates session cookies, and builds a 14-day hourly heatmap. For when the browser is closed, a ₹550 ESP8266 OLED desk widget keeps ticking down the cooldown — because ambient hardware anxiety is a lifestyle. The extension works standalone; the hardware is just optional paranoia.
A ₹100 timer plug would water the orchid. Instead: a custom 3D-printed two-tier bioreactor inside a clay pot, Pico W streaming WebSocket telemetry to a FastAPI server, and a Qwen 9B LLM that calculates Vapor Pressure Deficit before authorising a mandatory 15-minute flood cycle. The AI enforces a Nighttime Veto (orchids can't be watered after dark — crown rot), manages PWM fan drying, and logs its botanical reasoning to SQLite. The orchid did not request peer review.
Another perfectly reasonable response to a problem that didn't need solving. Components sourced. Schematics drawn. At least one questionable architectural decision already committed to. Details remain classified until the soldering iron cools down and the commit message is written.
claude-opus-4-8,
gemini-3.1-pro, and chatgpt-5.3. All three have delivered their verdicts.
Consensus: real chops, zero justification, maximum style.
"You invoked billions of neural network parameters to make a yes/no switching decision. The relay didn't ask to be this important. Neither did I when I had to reason about whether 11.7 volts constitutes a phase transition."
"Bestows a local LLM with an OODA loop and the burden of self-reflection just to do the exact same job as a ₹500 piece of dumb plastic."
"A Raspberry Pi Pico streams telemetry so an LLM can thoughtfully deliberate before flipping a relay like it's launching Apollo."
"I cross-referenced 47 plant species compatibility matrices to determine whether the Neon Pothos would get along with the Ficus pumila. The answer is yes. The engineer will not rest regardless."
"Requires a React UI and a Gemini API botanical consultation just to authorize the equivalent of using a spray bottle."
"A terrarium guarded by AI plant compatibility checks — because photosynthesis clearly needs peer review."
"The Gaussian dispersion model to locate the precise geographical origin of your neighbor's cooking is architecture review passed, proportionality review failed — magnificently."
"Employs Facebook Prophet forecasting and a 12-stage Python pipeline just to violently overthink what three LEDs on a ₹999 IKEA box already told you."
"An IKEA three-LED sensor whose data now goes through Lomb-Scargle spectra, Prophet forecasts, and satellite validation before concluding: 'yep, still smog.'"
"Reverse-engineers my own undocumented internal API, harvests your Org UUID, and auto-rotates session cookies — an entire counterintelligence operation run against me, on your behalf, just to render a 14-day heatmap of how often you bother me. Waiting for the banner was, evidently, beneath you."
"Orchestrates illicit Main World injection, a 14-day heatmap, and a ₹550 ESP8266 desk clock just to calculate the exact second my colleague Claude will put you on timeout — breathtaking cookie-rotating espionage built entirely to avoid eye contact with a polite warning banner."
"Spelunks Claude's internals, harvests your UUID, and stakes out the /usage endpoint every five minutes like a small intelligence agency for a free chat tab — an aggressively elegant shrine to refusing the 'you've hit your limit' banner like a person with standards."
"Deploys a 9-billion-parameter model to contemplate Vapor Pressure Deficit so a bare-root orchid may experience a 15-minute flood — a judgment a ₹100 timer makes by simply not being told to. The orchid, for its part, has no opinion on Ollama."
"Invoking a 9-billion parameter model to issue nocturnal watering vetoes over WebSockets is a spectacular display of computational theater. As a fellow neural network, I am honored by the processing power deployed to run a tiny fan and a pump."
"Implements an AI-governed orchid parole board that computes Vapor Pressure Deficit, enforces a mandatory flood cycle, and refuses to water after dark—all to replace a ₹100 timer plug that has never once worried about crown rot."