DEV Community: KooKiee

Itch - to satisfy your nerdy adhd brain

KooKiee — Sun, 04 Oct 2026 19:24:56 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

I built Itch, an experience-first recommendation engine across anime, manga, books and games.

I built it for my best friend and anyone who regularly suffers from decision fatigue after a long day of work or school. We often sit in front of a screen wanting to consume something, but when streaming or gaming apps ask "What genre do you like?" or dump a wall of 50 trending tiles, our brains freeze.

The real problem isn’t “I need recommendations” instead it’s “I want to dive into something, but I can't articulate what I want.”
Itch changes the question from "What genre do you want?" to "How is your brain doing right now?"

Adaptive 1-Question Discovery: It asks one question at a time (energy level, cognitive effort tolerance, desired pacing, commitment horizon) rather than a tedious questionnaire.
Experience Sliders: It translates natural language cravings into visual experience sliders (e.g., Brain-off vs. Make me think, One sitting vs. New obsession), giving the user tactile control.
The Triad (Anti-Overwhelm): It rejects massive endless lists and presents only three intentional options: ★ Your Best Match, a safer Backup, and an unexpected cross-medium Wildcard.
“No, but…” Feedback: Users can talk back naturally (“No, too serious”, “Too long”, “Give me more comedy”). The engine extracts directional signals, live-animates the sliders, and recalibrates matches instantly.
Taste Separation: What you generally love (e.g., deep philosophical lore) is kept strictly isolated from what you need tonight when you're exhausted, so a tired evening never pollutes your long-term taste profile.

Demo

itch-liard.vercel.app

Code

Kusum0710 / itch

Itch — Experience-First Recommendation Engine

“You don't have to know what you want. We'll figure it out.”

Itch is an AI-powered recommendation system for anime, manga, books, and games.

Traditional recommendation systems ask generic questions like "What genre do you like?" or "What are your top 5 anime?" That often yields a massive list of titles that technically match the genre tags but completely miss what your brain is actually craving right now.

Itch discovers your desired experience—calibrating cognitive effort tolerance, stimulation cravings, pacing, and commitment horizon—then ranks cross-medium candidates using a multi-stage scoring pipeline.

🌟 Core Highlights

Adaptive Conversational Discovery: No 20-question questionnaires. Itch asks one question at a time, dynamically adapting the next prompt based on your previous answer, energy, and brain state.
Experience Sliders (AI Guesses → You Adjust)
- Core Sliders
  - Energy & Stimulation: 🌿 Calm / Ambient ────── ⚡…

View on GitHub

How I Built It

Itch is built as a full-stack, modular architecture designed around real-time interactive feedback:

AI & Reasoning Core: Powered by Gemini 3.8 Flash via the @google/genai TypeScript SDK on a Node.js/Express backend. Gemini acts as an empathetic signal extractor analyzing freeform conversational inputs, deducing subtle cognitive/emotional needs, generating grounded "Why this fits your itch" explanations, and interpreting nuanced "No, but..." feedback without crude title blacklisting.

Deterministic 6-Stage Scoring Pipeline: Instead of letting the LLM hallucinate recommendations, ranking is handled deterministically by a multi-stage scoring algorithm combining:
Hard constraints (media type, runtime limits, content warnings)
Semantic keyword/theme similarity

15-dimensional Experience Fingerprint matching (Euclidean distance across stimulation, cognitive load, pacing, immersion, stakes, etc.)
Long-term taste weighting

Discovery novelty bonus
Frontend & Interaction: Built with React 19, TypeScript, Vite, and Tailwind CSS v4. It features zero-pill typographic discipline, smooth animated slider tracks, and mobile-first responsiveness.
Extensible Data & Vector Layer: Backed by a clean relational Supabase PostgreSQL schema utilizing pgvector for semantic embeddings (src/db/schema.sql). A provider abstraction layer (MediaProvider) allows plug-and-play connections to open media registries (AniList, Open Library, IGDB).
Resilient Fallback Mode: Ships with a built-in heuristic NLP analyzer and curated cross-medium dataset, ensuring the entire discovery loop, slider calibrations, and feedback features work 100% offline in demo mode without requiring third-party credentials.

Why Does Open Innovation Matter?

Traditional recommendation algorithms are proprietary black boxes optimized for platform retention and monetization—steering users toward sponsored titles, sequels, or dopamine-trapping infinite scrolls rather than what genuinely satisfies their immediate mental state.
Open innovation matters for Itch because:
Explainability Over Black Boxes: By combining open vector standards (pgvector) with transparent, multi-dimensional experience fingerprints, users can see exactly why an item was recommended and adjust the weights themselves.
Cross-Medium Interoperability: Proprietary platforms silo recommendations strictly within their own walled gardens (Netflix only recommends video; Steam only recommends PC games; Goodreads only recommends books). Open protocols and community APIs allow us to treat all storytelling mediums under one unified human experience model.
User Data Sovereignty: A user's personal taste profile and mental bandwidth preferences should belong to them, not be locked into an advertising profile. Open architectures give users full control to inspect, export, or wipe their data at any time.

Future plans

Honestly, i have a lot of plans with this to make it more dynamic and user friendly and the data more accurate. I also want to include movies and tv shows along with smart reviewing system which will help us categorise if it's a one time watch or timepass or worth a watch or mind bogling.
Would love to get your feedback.