Redesign homepage into personal showcase with project stories

Replace technical card descriptions with first-person project stories
(maintainer's voice, verbatim) and add italic section intros (ledes).
Soften the layout: larger readable story text, generous whitespace,
de-emphasized muted tech tags, subtle GitHub links under each story.
Keep dark mode + mobile friendly; Dockerfile and footer unchanged.

index.html

@@ -21,14 +21,14 @@
     <section class="section" id="cipherdolls">
       <div class="section-head">
         <h2 class="section-title">CipherDolls — AI companion platform</h2>
-        <p class="section-desc">An AI companion platform spanning cloud, web, and physical IoT devices.</p>
+        <p class="section-lede">What if an AI companion wasn't just an app, but something physical in your room? CipherDolls is my biggest project: an AI companion platform where characters can talk, listen — and live inside real devices.</p>
         <a class="section-link" href="https://github.com/cipherdolls" rel="noopener" target="_blank">github.com/cipherdolls ↗</a>
       </div>
       <div class="cards">
 
         <article class="card">
           <h3 class="card-name">core</h3>
-          <p class="card-desc">Backend API for the CipherDolls platform: wallet-signature auth, AI chat completions, TTS/STT, embeddings (pgvector), USDC payments on Base, and IoT device management over MQTT.</p>
+          <p class="card-story">The brain of the platform: one backend that handles chat, voice, memory, payments and the connection to the physical dolls.</p>
           <ul class="tags">
             <li>Bun</li><li>Elysia</li><li>Prisma</li><li>PostgreSQL</li><li>Redis</li><li>MQTT</li>
           </ul>
@@ -37,7 +37,7 @@
 
         <article class="card">
           <h3 class="card-name">server</h3>
-          <p class="card-desc">Minimal chat backend derived from core, stripped to the essentials: chat with a Character, optional TTS/STT — same architecture, no payments or IoT.</p>
+          <p class="card-story">A slimmed-down core for anyone who just wants to chat with a character — the same architecture, none of the complexity.</p>
           <ul class="tags">
             <li>Bun</li><li>Elysia</li><li>Prisma</li><li>PostgreSQL</li>
           </ul>
@@ -46,7 +46,7 @@
 
         <article class="card">
           <h3 class="card-name">webapp</h3>
-          <p class="card-desc">Web frontend for the platform.</p>
+          <p class="card-story">The web face of CipherDolls — where you meet the characters.</p>
           <ul class="tags">
             <li>React Router</li><li>TypeScript</li><li>Tailwind</li>
           </ul>
@@ -55,7 +55,7 @@
 
         <article class="card">
           <h3 class="card-name">stream-player</h3>
-          <p class="card-desc">TTS audio streaming proxy — subscribes to Redis Pub/Sub and forwards audio chunks to WebSocket clients with a compact binary protocol.</p>
+          <p class="card-story">Voices need to arrive instantly. This streams the AI's speech to your browser the moment it is generated.</p>
           <ul class="tags">
             <li>TypeScript</li><li>Redis</li><li>WebSocket</li>
           </ul>
@@ -64,7 +64,7 @@
 
         <article class="card">
           <h3 class="card-name">stream-recorder</h3>
-          <p class="card-desc">WebSocket audio ingestion microservice — accepts raw WAV streams, converts to MP3 in-memory via ffmpeg, forwards to the backend.</p>
+          <p class="card-story">The listening half: your voice streams in, gets converted on the fly, and lands with the AI.</p>
           <ul class="tags">
             <li>TypeScript</li><li>WebSocket</li><li>ffmpeg</li>
           </ul>
@@ -73,7 +73,7 @@
 
         <article class="card">
           <h3 class="card-name">firmware-watcher</h3>
-          <p class="card-desc">ESP-IDF firmware for the SenseCAP Watcher companion device: BLE WiFi provisioning, MQTT events, MP3 playback through the onboard codec, LVGL UI on a 412×412 display, deep sleep.</p>
+          <p class="card-story">The magic moment: a real device on your desk wakes up, connects, and speaks with the character's voice.</p>
           <ul class="tags">
             <li>C</li><li>ESP-IDF</li><li>ESP32-S3</li><li>LVGL</li>
           </ul>
@@ -82,7 +82,7 @@
 
         <article class="card">
           <h3 class="card-name">firmware-m5sticks3</h3>
-          <p class="card-desc">ESP-IDF firmware for the M5StickS3 (ESP32-S3) pocket companion device with display, audio codec, and IMU.</p>
+          <p class="card-story">The pocket version: the same companion on a tiny stick with a display and a microphone.</p>
           <ul class="tags">
             <li>C</li><li>ESP-IDF</li><li>ESP32-S3</li>
           </ul>
@@ -96,13 +96,13 @@
     <section class="section" id="dapps">
       <div class="section-head">
         <h2 class="section-title">Decentralized apps — Gnosis Chain + Ethereum Swarm</h2>
-        <p class="section-desc">dApps with on-chain logic and censorship-resistant Swarm storage/hosting.</p>
+        <p class="section-lede">Can the web run without servers? These apps are my experiments in building things that nobody can switch off.</p>
       </div>
       <div class="cards">
 
         <article class="card">
           <h3 class="card-name">SwarmChat</h3>
-          <p class="card-desc">Decentralized 1:1 messenger and WebRTC video calling: wallet identity, on-chain contact registry, encrypted messages over Swarm PSS with feeds as offline outbox. Live at swarmchat.eth.</p>
+          <p class="card-story">I was thinking about how cool it would be to have a web3-based chat system. Is it possible to communicate without any servers? With Swarm, a decentralised technology, this is now possible — messages travel encrypted through the network itself, and your wallet is your identity.</p>
           <ul class="tags">
             <li>Solidity</li><li>React</li><li>Swarm PSS</li><li>WebRTC</li>
           </ul>
@@ -111,7 +111,7 @@
 
         <article class="card">
           <h3 class="card-name">Pink Chainsaw</h3>
-          <p class="card-desc">Decentralized imageboard: post images, comment, and vote with xBZZ; fees auto-top-up the poster's Swarm postage stamp so good content stays alive. A social-score system prices posting by reputation.</p>
+          <p class="card-story">I wanted an imageboard where content can't just disappear. Images live on Swarm, and every post pays a little to keep them alive — good content earns you cheaper posting, bad content gets expensive.</p>
           <ul class="tags">
             <li>Solidity</li><li>Foundry</li><li>React</li><li>Swarm</li>
           </ul>
@@ -120,7 +120,7 @@
 
         <article class="card">
           <h3 class="card-name">FreeeMarket</h3>
-          <p class="card-desc">Multi-vendor marketplace where anyone runs their own shop and gets paid in stablecoins through on-chain escrow — deliberately fee-free and VAT-free.</p>
+          <p class="card-story">Can you run a marketplace with no company behind it? Anyone opens a shop, buyers pay in stablecoins, and on-chain escrow replaces the middleman. Deliberately fee-free — a farmers' market for the internet.</p>
           <ul class="tags">
             <li>Solidity</li><li>Gnosis Chain</li><li>Swarm</li>
           </ul>
@@ -129,7 +129,7 @@
 
         <article class="card">
           <h3 class="card-name">WalletMatrix</h3>
-          <p class="card-desc">Pivot-table view of a crypto wallet across every EVM chain — tokens as rows, chains as columns; drag one cell onto another to bridge or cross-chain swap. Static SPA deployable to Swarm.</p>
+          <p class="card-story">My tokens were scattered across a dozen chains and I could never see the whole picture. So I built a pivot table for your wallet: tokens as rows, chains as columns — drag a cell from one chain to another to bridge.</p>
           <ul class="tags">
             <li>React</li><li>TypeScript</li><li>Vite</li>
           </ul>
@@ -138,7 +138,7 @@
 
         <article class="card">
           <h3 class="card-name">swarm-connect</h3>
-          <p class="card-desc">React connect-button + wizard for Swarm on Gnosis: wallet → network → balances → Bee node → postage stamp, in one modal. Published on npm; demo hosted on Swarm itself.</p>
+          <p class="card-story">Every Swarm dApp I built needed the same onboarding dance: wallet, network, Bee node, postage stamp. So I packaged it once — a single React button that walks users through all of it.</p>
           <ul class="tags">
             <li>React</li><li>wagmi</li><li>npm package</li>
           </ul>
@@ -147,7 +147,7 @@
 
         <article class="card">
           <h3 class="card-name">opentab</h3>
-          <p class="card-desc">Onchain app bootstrapped with create-onchain (Base).</p>
+          <p class="card-story">An experiment with onchain apps on Base.</p>
           <ul class="tags">
             <li>Next.js</li><li>TypeScript</li>
           </ul>
@@ -161,13 +161,13 @@
     <section class="section" id="ai-gpu">
       <div class="section-head">
         <h2 class="section-title">AI on GPU — self-hosted speech services</h2>
-        <p class="section-desc">Dockerized speech AI for NVIDIA GPUs, tuned for the RTX 5090 / Blackwell generation.</p>
+        <p class="section-lede">I wanted my projects to hear and speak without sending a single word to the cloud — so these run on my own GPU.</p>
       </div>
       <div class="cards">
 
         <article class="card">
           <h3 class="card-name">whisper-cuda</h3>
-          <p class="card-desc">OpenAI-Whisper-compatible ASR web service built for RTX 5090 (sm_120) — faster-whisper on CTranslate2/CUDA 12.8, no PyTorch.</p>
+          <p class="card-story">Speech-to-text on my own GPU: no cloud, no per-minute pricing. Whisper packaged for the RTX 5090 so anything in my homelab can transcribe audio locally.</p>
           <ul class="tags">
             <li>Python</li><li>FastAPI</li><li>CUDA</li>
           </ul>
@@ -176,7 +176,7 @@
 
         <article class="card">
           <h3 class="card-name">kokoro-cuda</h3>
-          <p class="card-desc">Minimal Kokoro TTS API with CUDA: 49 voices in 9 languages, streaming WAV/PCM plus MP3/Opus/FLAC, web UI included.</p>
+          <p class="card-story">Giving my projects a voice: a small self-hosted TTS service with 49 voices in 9 languages, fast enough to stream. Type text, hear it instantly.</p>
           <ul class="tags">
             <li>Python</li><li>FastAPI</li><li>CUDA</li>
           </ul>
@@ -185,7 +185,7 @@
 
         <article class="card">
           <h3 class="card-name">voxtral-cuda</h3>
-          <p class="card-desc">Text-to-speech service running Mistral's Voxtral-4B-TTS with CUDA acceleration via vLLM-Omni.</p>
+          <p class="card-story">The newest open speech models are amazing but heavy. This runs Mistral's Voxtral on a local GPU, so state-of-the-art speech stays in my own house.</p>
           <ul class="tags">
             <li>Python</li><li>vLLM</li><li>CUDA</li>
           </ul>
@@ -204,7 +204,7 @@
 
         <article class="card">
           <h3 class="card-name">stakingtracker</h3>
-          <p class="card-desc">Self-hosted ETH staking rewards tracker and tax-report generator — reads from your own beacon node, persists daily history to SQLite, no external APIs or keys.</p>
+          <p class="card-story">Tax season with an ETH validator used to mean paid APIs and spreadsheets. Now my own beacon node does the work: a small self-hosted app that records rewards daily and prints the tax report.</p>
           <ul class="tags">
             <li>TypeScript</li><li>Fastify</li><li>SQLite</li>
           </ul>