AI Workflows & Experiments by Darwin Hernandez

Python / Music Intelligence Experiment

Crate Digger

Finds music that matches your mood, not just what an algorithm already knows.

Crate Digger helps listeners describe a mood, moment, or sonic direction and discover albums that fit. Instead of recycling popularity signals, it calibrates context, genre distance, and artist visibility, then returns evidence-backed recommendations with clear confidence and honest limits.

What this proves

Music discovery can be more personal and trustworthy when context, not popularity alone, guides the path and every recommendation shows why it fits.

01 Context-led discoveryStarts with a listener's mood, moment, or sonic direction, then translates it into a focused path to music that fits.

02 Evidence you can inspectExplains why each album matches the request while keeping verified identity, curator judgment, third-party signals, and model inference distinct.

03 Honest coverage statesSays when the available evidence is too weak, so discovery does not become another confident algorithmic guess.

Next.js / Sports Analysis Experiment

OctoOracle

A playful World Cup prediction app powered by deterministic match logic and an animated octopus.

A World Cup 2026 experience where a playful animated octopus reveals a source-backed prediction for the featured match. The app combines fixture data, structured statistical signals, public-source research, prediction history, and an animated aquarium reveal without presenting outcomes as certainty.

What this proves

Entertainment apps can still be thoughtful systems: transparent, repeatable, source-aware, and honest about uncertainty.

01 Deterministic predictionsGenerates reproducible probabilities, expected goals, scorelines, confidence, and public explanations.

02 Animated aquarium revealUses a playful octopus character to make the prediction feel memorable, repeatable, and fun.

03 Built with safeguardsFreezes predictions at kickoff, tracks results, handles stale data honestly, and avoids presenting predictions as betting advice.

Next.js / AI Workflow Experiment

Business Lead Assistant

Turns vague contractor inquiries into clear, qualified project briefs.

A multi-tenant lead intake app for local service businesses, starting with renovation and home-service contractors. Each contractor gets a branded intake page where homeowners submit project details, photos, budget range, timeline, and contact preferences. The system then uses AI to summarize the lead, score fit, flag missing information, and notify both contractor and homeowner.

What this proves

AI is most useful when it turns messy real-world inputs into structured decisions people can act on.

01 Guided intakeCaptures project type, goals, location, timeline, budget range, photos, and preferred contact method.

02 AI lead qualificationGenerates a plain-language summary, fit score, missing information, and recommended next step without giving unsafe construction, permit, cost, or legal advice.

03 Small-business workflowSupports contractor notifications, homeowner confirmations, lead photo uploads, multi-tenant storage, and an admin dashboard for reviewing and managing leads.

Python Experiment

music-crewai

Turns any music genre into a multi-file research report, saved per topic and viewable in a Gradio UI.

A research assistant for music scenes. Give it a genre and it organizes the history, artists, context, and patterns into a readable report.

What this proves

Agent workflows can turn curiosity into structured output that people can actually use.

01 Question, research, write, package, explore.

02 Built to automate deep research and make it explorable.

03 Open source. Always iterating.

TypeScript Experiment

Puchi & Pao's Sparkling Adventure

A local Phaser MVP for a 16-bit-inspired 2D platform adventure.

A playable Level 1 game prototype with title, story, character select, gameplay, game over, and level complete screens.

What this proves

Games are a direct way to practice product feel: timing, feedback, state, constraints, and delight.

01 Puchi and Pao move differently, creating real character feel.

02 Stars, shells, crabs, water zones, powers, checkpoints, and a Giant Clam Shrine goal make the loop playable.

03 Built with generated placeholder art, particles, chiptune music, and Web Audio sound effects.

Python Experiment

family-planner

An organizer for family events, tasks, schedules, and shared lists.

A personal operations tool for coordinating real life: plans, responsibilities, reminders, and household context.

What this proves

Automation is most meaningful when it supports the relationships and routines people already care about.

01 Good systems make care easier to practice.

02 Automation is most meaningful when it supports the relationships and routines people already care about.

Python Experiment

job_search_agent

An agent workflow that helps prioritize job posts, tailor resumes, and monitor applications.

A practical AI workflow for making a high-friction search process more organized, personalized, and trackable.

What this proves

Agents are most helpful when they reduce repeated judgment-heavy admin work.

01 Use AI to focus attention where decisions matter.

02 Agents are most helpful when they reduce repeated judgment-heavy admin work.

TypeScript Experiment

kanban_board

A simple Kanban board to visualize tasks, track progress, and stay focused on what matters.

A lightweight planning surface for moving work from idea to done without drowning in process.

What this proves

Useful product thinking starts with making the state of work visible.

01 Clarity is a productivity feature.

02 Useful product thinking starts with making the state of work visible.

Ideas in code. Curiosity in motion.