Failed Projects
A Whatsapp bot to handle advertisements on the app, with automatic billing for group members for a selected time period Reason : WhatsApp API was too troublesome to set up + costly for an MVP
Automatic trading strategy generation from natural language to python code through a DSL. Reason : A lot of it could just be a simple n8n workflow, and a DSL was overkill. Although I think a simple injection of generated code (the strategy) into a ready-made framework can work pretty well.
Monetary tracking of AI workflows per project (An AI project manager ?) Reason : Infeasible to create my own AI workflow generator, especially since n8n has hundreds of integrations. Could add custom tracking over n8n, but that requires deeper changes in n8n, and only on custom machines (not commercial).
Mock data generator for drizzle schemas Reason : Did not know that drizzle-seed exists (basically just started the project, didn't bother for similar projects lol).
Get manga cut-outs, and animate them (basically manga animations). Reason : Requires multiple deep learning models, may have been above my then programming and hardware level. (Though I think it can be feasible now, damn gonna try it out).
CSV files data interpretation pipeline Reason : Ingest a CSV file, get an LLM to pick necessary columns needed for training classical ML models, train them, interpret the models using libraries, feed the results back into an LLM, and ask it for a meaningful interpretation. This does sound like a good "AutoML" system, only the idea was a little pre-mature, and was made before the models got better.