Blog on Chengyu Wang

Building an Automatic Podcast Transcription and AI-Organizing System on My Synology NAS

Tue, 18 Aug 2026 18:00:12 +0000

Rather than trying to run Whisper locally on an underpowered NAS, I offloaded transcription to Cloudflare Workers AI and article cleanup to LiteLLM, turning the NAS into an automation hub with its own web player for listening, searching, and reading.

How Does a Machine "Figure Out" Something You Never Told It? A Look at Ontological Reasoning

Fri, 14 Aug 2026 16:00:24 +0000

Databases only know what you explicitly store; ontological reasoning lets a machine derive thousands of unstated facts from a handful of concept definitions — powerful, expensive to build, and increasingly paired with LLMs to keep each other’s weaknesses in check.

DSM 7.4 Upgrade Log: A Smooth Update With One Small Letdown

Mon, 20 Jul 2026 08:00:54 +0000

Community reports convinced me DSM 7.4 was stable enough to risk on a production NAS — the upgrade went flawlessly, but the one feature I actually wanted, Storage Efficiency, turned out to be locked out of older hardware entirely.

Tinkering Log: Getting New Tricks Out of a DJI 4G Module

Fri, 10 Jul 2026 07:55:28 +0000

A DJI 4G module identifies itself with a private USB ID that Linux won’t drive properly — spoofing it as a generic Quectel EC25 unlocked full modem support, and it ended up running SMS management on an OpenWrt router instead of the NAS.

Immich 3.0 Upgrade Log: Taking a DS218+ From 2.7 to 3.0

Thu, 09 Jul 2026 18:00:12 +0000

The official 3.0 release with built-in AI photo editing was tempting enough to risk an upgrade on a stable NAS setup — the process went smoothly, with the only surprise being a slow first-boot database migration that looked like a crash.

Tracking Down a Podsync 'Empty Duration' Crash: From a Bad Format String to Filling In the PO Token Gap

Thu, 09 Jul 2026 14:08:34 +0000

One feed kept crashing Podsync with a cryptic ‘bad format string’ error — three wrong turns later, the real fix meant switching Docker images and realizing YouTube’s PO Token wasn’t actually being generated at all.

From a Compromised VPS to Building a Real Website Backup System

Fri, 03 Jul 2026 10:00:10 +0000

After cleaning up malware off a VPS, the bigger realization was that defense alone isn’t enough without recoverability — so I built a four-layer backup system covering local snapshots, scheduled automation, off-site NAS sync, and recovery documentation.

Reflections After a VPS Security Incident: From Discovery to Building a Full Monitoring System

Thu, 02 Jul 2026 12:37:21 +0000

A cryptomining-style trojan ran undetected on my Oracle Cloud VPS for nearly 20 days — the real lesson wasn’t fixing that one box, but realizing monitoring itself needs to be monitored.

From Dropbox to Seafile: Letting the Files on My NAS Slowly Grow Into a Knowledge Base

Mon, 15 Jun 2026 08:00:48 +0000

After bringing photos home to Immich, files were next — Seafile replaced Dropbox for syncing, then got paired with Open WebUI, RAG, and LiteLLM to turn a plain file store into a searchable, AI-queryable knowledge base.

From Google Photos and iCloud to Immich: Building My Own Photo Management Hub

Mon, 15 Jun 2026 08:00:36 +0000

Photos scattered across Google Photos, iCloud, a Huawei phone, a DJI drone, and action cameras finally got consolidated into one self-hosted, deduplicated Immich library — though my aging Synology DS218+ is starting to show its limits.

AutoReel: A NAS-Oriented Tool for Auto-Organizing Movie and TV Files

Sun, 24 May 2026 22:00:08 +0000

Built a small self-hosted tool that watches a download folder, identifies movies and TV episodes, and renames/moves them into an Emby/Jellyfin/Plex-friendly structure — deliberately conservative, preferring to flag unrecognized files rather than guess and rename wrong.

Podsync Monitoring and Alerts: Catching Failures Before Episodes Stop Updating

Tue, 05 May 2026 18:00:08 +0000

Self-hosted Podsync fails silently when YouTube cookies get rotated out — a simple log-scanning script paired with scheduled tasks now pushes an alert within 10 minutes, plus twice-daily download reports.

Gauging Coding Models' Real Reputation Through Hacker News

Mon, 04 May 2026 22:00:09 +0000

A site that runs AI sentiment analysis on real Hacker News developer comments, rather than benchmark scores — Claude comes out on top for engineering sense, GPT close behind, with DeepSeek a notable dark horse.

Building My Own Information Feed Hub With TrendRadar

Mon, 04 May 2026 21:00:13 +0000

Too many English finance and tech RSS feeds to skim by hand — TrendRadar, an open-source aggregator, filters, translates, and summarizes them via AI, then pushes the results to Telegram.

Fixing a Broken Cover Image in My Podcast RSS Feeds

Mon, 04 May 2026 20:00:08 +0000

One podcast pipeline’s cover art broke when its external image URL died; the fix was routing RSS cover URLs through a single local-file method, plus a couple of quality-of-life tweaks for a language-learning audio feed.

How I Rebuilt My Hermes Assistant So My Life's Domains Stop Bleeding Into Each Other

Wed, 29 Apr 2026 10:00:06 +0000

Investing talk kept leaking into work notes and vice versa — the fix was splitting Hermes into a router plus dedicated stock, work, diary, and life sub-assistants, each with its own knowledge base partition and a strict no-cross-referencing prompt.

Podsync Download Troubleshooting Diary

Wed, 29 Apr 2026 09:00:07 +0000

Podsync started throwing YouTube bot-verification and rate-limit errors — the fix was exporting browser cookies into the container’s mounted data directory and wiring them into yt-dlp’s args.

ChatGPT Images 2.0 Launched — Here's How It Performs

Thu, 23 Apr 2026 22:00:19 +0000

Four quick tests of the new ChatGPT Images 2.0: visualizing my current life as the AI understands it, an imagined retirement, a comic-strip take on a 2013 journal entry about orchids, and a PowerPoint-generation test.

This Weekend I Finally Built the Three-Part AI System I'd Been Thinking About

Mon, 20 Apr 2026 02:01:20 +0000

A personal assistant on the home NAS, a home assistant tied into an investment system, and an intelligence-gathering agent on a VPS — three modules starting to work together as a real perception-cognition-action loop, not just a pile of tools.

Getting LiteLLM's Model Routing Set Up Properly

Sun, 19 Apr 2026 23:00:11 +0000

A full writeup of the LiteLLM gateway config behind my AI stack — free NVIDIA NIM and OpenRouter models as the safety net, cheap paid models like DeepSeek V3 for daily use, and Claude Sonnet as an on-demand top-tier fallback.

Migrating Hermes to My Synology NAS

Sun, 19 Apr 2026 22:00:12 +0000

After getting Hermes running on a VPS, I moved it to my Synology NAS instead so it could gradually gain access to household journal and investment data as an internal AI hub — and redesigned the model routing along the way.

Hermes: Setting Up My New AI Assistant

Sun, 19 Apr 2026 01:00:07 +0000

OpenClaw’s buzz is fading and everyone’s talking about Hermes — its token compression alone (50%+, reportedly up to 70% in some cases) was reason enough to spin up a VPS and try it.

Testing Models Against a Real Use Case, Not Just Benchmarks

Sat, 18 Apr 2026 17:00:16 +0000

Blind-testing two Ollama models against real CIPLE A2 Portuguese-exam questions, the well-documented, benchmark-topping model lost badly to an obscure community fine-tune — because the benchmarks were measuring Brazilian Portuguese, not European.

Build From What Already Exists, Then Generate: An AI Audit-Rules Project

Sat, 18 Apr 2026 10:01:41 +0000

Writing the plan for an AI-generated audit-rules project, we chose to reverse-engineer rules from years of proven production scripts rather than rebuilding a rule system from raw data and business logic.

VibeVoice Setup Diary (ASR + TTS Exploration)

Sat, 18 Apr 2026 10:01:18 +0000

Setting up Microsoft’s VibeVoice to test audio-to-subtitle generation and text-to-speech: the ASR side (multi-speaker SRT generation) turned out production-ready, while TTS — especially Chinese — wasn’t there yet.

Tracking Down Why My Ubuntu Server Kept Auto-Restarting After Shutdown, Post Wake-on-LAN Setup

Sat, 18 Apr 2026 09:00:08 +0000

A GPU server configured for Wake-on-LAN would shut down cleanly, then reboot itself ~10 seconds later — the culprit was ACPI wakeup left enabled on the network card, letting ordinary LAN broadcast traffic trigger a wake, not just the magic packet.

RemNote's iPhone Flashcard Text Too Small? Fixed It With CSS

Tue, 14 Apr 2026 10:28:54 +0000

RemNote’s iOS app has no font-size setting for flashcards — after confirming custom CSS actually gets through the WebView, a couple of targeted selectors made the card text readable without touching the rest of the UI.

OpenClaw's Weekend "Zero-Cost" Overhaul

Sat, 11 Apr 2026 17:33:00 +0000

When my free Google Cloud credits ran out and Gemini started rate-limiting, I rebuilt OpenClaw’s model chain into a three-tier free-and-cheap failover setup — and rethought which use cases were still worth the tokens.

Breaking Free: Batch-Migrating OnePlus Notes With Claude in Chrome

Sun, 15 Mar 2026 14:45:00 +0000

When ad-stuffed ‘phone switching’ apps couldn’t move 352 notes to a new Huawei phone, reverse-engineering the OnePlus cloud web app’s API and scripting it through Claude’s Chrome extension got the job done in 20 minutes.

Optimizing OpenClaw's Quiz Response Speed

Sun, 15 Mar 2026 00:04:00 +0000

Adding a quiz question was taking 2-3 minutes and getting slower over time — the fix was switching the data file from a JSON array to NDJSON so new items append in one line instead of a full rewrite.

Chasing Down OpenClaw's Slow-Reply Problem: Auth Profiles and Session Stickiness

Sat, 14 Mar 2026 22:00:47 +0000

Adding more API keys didn’t fix the slowdowns — the real issue was a key that existed in secrets but was never registered for failover, plus session pinning that kept using the same key after a restart.

Is OpenClaw Actually Worth Installing?

Thu, 12 Mar 2026 09:00:22 +0000

An honest review after months of daily use, from when it was still called ClawdBot: genuinely useful for memory, reminders, and lightweight investing research — but with real drift and security caveats.

Trying Out Claude's Finance Analysis Feature on NVDA

Thu, 26 Feb 2026 18:00:24 +0000

Setting up Claude’s Finance Analysis tool through Cowork, then having it produce a full comps-based valuation report on NVIDIA against AMD, Broadcom, Qualcomm, Marvell, and Intel.

NVIDIA's Blackwell Architecture, Explained

Thu, 26 Feb 2026 06:00:06 +0000

Two full-reticle dies fused into one GPU, FP4 inference, and a fifth-generation NVLink that connects up to 576 GPUs — the architecture behind NVIDIA’s ‘AI factory’ strategy.

Idea to Blog, WeChat, and Short Video in One Shot With OpenClaw + macOS

Thu, 26 Feb 2026 05:00:35 +0000

Moving OpenClaw off a Linux VPS and onto a hackintosh, then adding NotebookLM’s CLI to turn a finished blog post into a narrated slide video automatically.

The Real Variable in the AI Era Isn't a Recession — It's Wealth Redistribution

Wed, 25 Feb 2026 11:41:00 +0000

Reacting to Citrini Research’s viral 2028 doom scenario: technological revolutions have never shrunk the economy, but the gap between people who can afford frontier AI and those who can’t is opening faster than any previous divide.

OpenClaw × Blog × WeChat: Wiring Up a Full Automation Pipeline

Mon, 23 Feb 2026 10:00:08 +0000

From a rough idea in OpenClaw to a published blog post to a WeChat Official Account draft — with a VPS script standing in as the bridge and quietly fixing WeChat’s image hotlink problem along the way.

Configuring Blogwatcher, for a Wider View

Thu, 12 Feb 2026 15:00:10 +0000

Watching an AI-curated information feed surface a YC ’ex technical founder’ listing made one thing click: the real skill now is building systems with AI, not just using it.

Connecting Clawdbot to Obsidian, Apple Notes, and Everything Else

Thu, 12 Feb 2026 11:00:12 +0000

Voice in, an Obsidian note out, a blog post and short video out the other end automatically — plus Apple Notes, Google Sheets, and an email pipeline running hands-free.

Three Models, All Working for Me Now

Wed, 11 Feb 2026 09:00:07 +0000

Codex writes the code fast, Claude Code audits it, Gemini plugs into OpenClaw as a tool, NotebookLM organizes notes, and Nanobanana handles visuals — a small AI production line.

Weekend Chaos: Auto-Login Broke My Hackintosh, and AI Talked Me Back In

Mon, 09 Feb 2026 16:00:12 +0000

One innocent auto-login setting later, my Mac mini hackintosh was stuck in a boot loop — a night of Gemini-guided EFI edits, a networkless recovery dead end, and a reinstall that saved my data.

Programming Is Just Typing — Your Experience Is the Real Asset

Sun, 08 Feb 2026 14:00:29 +0000

My notes on Jensen Huang’s unusually candid remarks at a 2026 AI summit: syntax stopped being the moat, and domain expertise became the whole game.

AI Benchmarks: Opus 4.6 vs. GPT-5.2 vs. Gemini 3 Pro, Head to Head

Sun, 08 Feb 2026 14:00:22 +0000

A benchmark chart making the rounds shows large models shifting from chatbots to agents — with Opus 4.6 dominating agentic computer use, GPT-5.2 still king of pure reasoning, and Gemini 3 Pro owning vision and multilingual work.

"Code Is Dead"? What Building a Platform and App Solo With AI Actually Taught Me

Sun, 08 Feb 2026 12:00:11 +0000

Codex framed the app, Cursor did the surgical edits, and a working SIP call-center app got built by one person — the rough UI is the actual lesson here.

Deleting My Fine-Tuning Notes: From 'Taming a Model' to 'Steering a Process'

Sun, 08 Feb 2026 03:00:07 +0000

Clearing out a year-old Obsidian folder on model fine-tuning made the shift obvious: from chasing model capability to chasing business results through workflow and RAG.

Tech Log: Chasing Down a Hijacked LAN Route on macOS

Thu, 05 Feb 2026 11:00:06 +0000

A NAS on 192.168.5.x went unreachable with ‘No route to host’ — the culprit was a leftover feth interface from an ad-blocker’s content-filter extension that never fully uninstalled.

My AI Content Workflow: From Logic Outline to a High-Quality Deck

Wed, 04 Feb 2026 23:00:38 +0000

A three-step pipeline — outline with Gemini, flesh out a transcript, then let NotebookLM turn both into a presentation — that cut a week of course prep down to a day or two.

Ops Diary: Expanding My Hyonix VPS

Wed, 04 Feb 2026 22:00:21 +0000

A WSL2 dead end on an old Windows Server 1809 box turned into a pleasant surprise: one support ticket doubled the disk quota on a four-year-old VPS.

Two Days Deep in Moltbot: AI Is Reshaping the Deploy-Configure-Use Path

Fri, 30 Jan 2026 11:00:21 +0000

Watching an agent diagnose its own missed schedule, switch itself from passive to active wake mode, and confirm the fix — without me tracing a single step of it myself.

A Failed Attempt: Tracking a Google Account Suspension and Appeal

Thu, 29 Jan 2026 15:00:09 +0000

Registered a new Google account to pay for steadier API access after hitting rate limits on the free tier — and got flagged as bot-created within a day of paying.

Forcing HTTPS on My Typecho Blog on Tencent Cloud Lighthouse

Wed, 28 Jan 2026 22:20:13 +0000

The blog answered on both HTTP and HTTPS at once — tracking down the default_server catching every HTTP request and rewriting it into a clean 301 chain to a single canonical domain.

Clawdbot 2026: VPS Deployment and Hardening Diary

Wed, 28 Jan 2026 21:59:54 +0000

Working around a broken systemd installer, teaching a model to fetch stock data itself when the plugin store isn’t reachable, and locking the firewall down to SSH-only.

Today's Ops Log: Switching AI Tools and Reworking the Network

Fri, 23 Jan 2026 15:00:18 +0000

Moving from Cursor to OpenAI Codex, two days spent simplifying the network layout, a fix for Podsync’s YouTube rate-limiting, and a disposable-container trick for benchmarking a NAS.

Fully Uninstalling Squirrel (Rime) on macOS

Thu, 22 Jan 2026 12:00:27 +0000

Going back to native macOS input plus voice dictation, and the full teardown needed since deleting the app alone leaves the input method partially resident.

Making Caddy Reverse-Proxy FreePBX Alongside Another Internal Web Service

Wed, 21 Jan 2026 23:00:24 +0000

FreePBX kept redirecting the browser back to its raw :8080 port even behind a reverse proxy — the fix was rewriting the Location header on the way out, not just spoofing headers going in.

Switching My Blog to a Free ACME Certificate

Wed, 21 Jan 2026 17:00:07 +0000

Tired of Tencent Cloud’s free certificate silently expiring and needing manual re-binding, I moved my Typecho blog over to acme.sh via a Gitee mirror instead.

From Dragging Nodes to Stating Intent: Claude Skills Is Ending the Old AI Workflow Era

Wed, 21 Jan 2026 10:00:12 +0000

Why describing what you want in plain language and letting Claude auto-apply a Skill beats another afternoon spent wiring up nodes in Dify, n8n, or Coze.

FreePBX Deployment Diary #3: The First Successful Test Call

Mon, 19 Jan 2026 00:01:07 +0000

Creating extensions, getting UDP registration working first, then chasing down a misleading ‘wrong password’ error that was actually a TLS transport and certificate mismatch.

FreePBX Deployment Diary #2: HTTPS and Automatic Backups

Sun, 18 Jan 2026 20:00:47 +0000

Getting Let’s Encrypt working on the admin panel and setting up a weekly automatic backup — the two unglamorous steps that make a FreePBX box actually production-ready.

TPU vs. GPU: A Deep Dive, and Who's Actually Using TPUs

Sun, 18 Jan 2026 17:00:08 +0000

GPUs are the flexible all-rounder with a mature ecosystem; TPUs are the specialist that’s brutally efficient at large-scale matrix math — plus a rundown of who actually trains on TPUs.

What Two AI Models Concluded From My Accident's Dashcam Footage

Sun, 18 Jan 2026 15:01:38 +0000

Asking Gemini Pro and ChatGPT 5.2 to do an engineering-style read on whether emergency braking was slow to react — and what the physics of a ‘ghost probe’ collision actually allow.

FreePBX Deployment Diary #1: Install and Network Setup

Sun, 18 Jan 2026 00:01:12 +0000

First contact with FreePBX 17 on Debian — a Vultr trial instance, the official one-shot install script, and the firewall gotcha that can lock you out of your own server.

Head-to-Head: How Mainstream AI Models Judge a Real Traffic Accident

Thu, 15 Jan 2026 18:51:39 +0000

After my own Tesla was hit by a car pulling out from under a bridge, I ran the accident photo past ChatGPT, Gemini, Grok, Copilot, Claude, and a few domestic models to see which ones actually got liability right.

When AI Starts Thinking Like a Designer: Comparing Gemini, Claude Opus, and GPT on a Real Redesign

Wed, 07 Jan 2026 12:00:14 +0000

Three models, one vague brief to redesign a blog’s UX, visuals, and SEO — and a clear split between models that plan first and models that just ship a result.

Looking Again at NVIDIA's Real Logic After CES 2026

Wed, 07 Jan 2026 12:00:00 +0000

Jensen Huang’s keynote wasn’t really about FLOPS — it was about token cost, and NVIDIA quietly shifting from selling chips to selling the entire stack.

Veo 3.1 vs. Jimeng: Stress-Testing an Image-to-Video AI Workflow

Thu, 18 Dec 2025 18:02:01 +0000

The same prompt, two video models — one produced a slightly rough transition clip, the other gave a character a 180-degree head spin.

Setting Up Squirrel + Rime-Ice: My New macOS Input Method

Thu, 18 Dec 2025 18:00:36 +0000

Native-smooth typing, a dictionary that actually keeps up with slang, and built-in emoji/spacing niceties — Rime finally out of the box.

Registering an Indian Apple ID, the Roundabout Way

Thu, 18 Dec 2025 10:00:19 +0000

Registering an Indian Apple ID directly kept failing — the fix was registering a Chinese account with no proxy first, then switching region afterward.

A Simple, Solid macOS Cleanup Tool: Mole

Tue, 16 Dec 2025 07:00:36 +0000

A restrained, transparent command-line cleaner for leftover app files and system/dev caches — scan first, then clean, nothing automatic or hidden.

First Day With GPT-5, and How AI Has Changed My Daily Life

Fri, 08 Aug 2025 17:00:09 +0000

From debugging code to language practice to research, ChatGPT has quietly become a daily-use tool rather than a novelty.

Fixing an Exposed Directory Listing in My Podsync Docker Service

Wed, 09 Jul 2025 11:00:06 +0000

A missing index.html meant Podsync’s web server was serving up a full directory listing of everything it had synced — here’s the one-file fix.

100 Projects Before I Retire, #6: A Real-Time Refreshing Asset Dashboard

Wed, 09 Jul 2025 10:00:11 +0000

A small web app for tracking multi-currency assets, with live exchange rates, category and currency breakdown charts, and inline editing.

100 Projects Before I Retire, #5: A Language-Learning App Built on Nursery Rhymes and Fairy Tales

Tue, 17 Jun 2025 19:00:08 +0000

Frustrated by bloated, ad-filled kids’ language apps, I built a clean native Android app that loads curated foreign-language rhymes and stories through a WebView.

100 Projects Before I Retire, #4: A MinIO Object Storage Service

Mon, 16 Jun 2025 09:00:05 +0000

Turning a cheap, low-spec server into a dedicated S3-compatible object store for images, audio, and video, instead of hosting another website on it.

Agents vs. Workflows: What's the Actual Difference?

Fri, 13 Jun 2025 16:00:16 +0000

A drag-and-drop OCR-plus-LLM tool got me thinking about where a workflow ends and an agent begins — with a simple three-question test.

Swapping My Parents' ISP Router for a Spare Xiaomi Router (an AP-to-Router-Mode Story)

Wed, 11 Jun 2025 11:00:20 +0000

A router that stubbornly stayed in AP mode after being moved, and the DHCP debugging it took to switch it back.

Reviving a Shelved Router: Reflashing a GL-SFT1200

Wed, 11 Jun 2025 11:00:16 +0000

A painfully slow GL.iNet Opal, rescued from a drawer with a two-step official firmware upgrade instead of a U-Boot flash.

Building a Smart Customer-Service Agent with Qwen-Agent

Wed, 11 Jun 2025 11:00:12 +0000

A telecom customer-service example — using Tools for live data lookups and a knowledge base for static rules, with an LLM deciding which to call.

Debugging a Broken Homepage: When a .mov File Got Treated as a Cover Image

Tue, 10 Jun 2025 16:00:09 +0000

A cover-image extractor that grabbed the first tag without checking it was actually an image — and broke the homepage whenever a post embedded video.

A Night With Google's New Video-Generation Model

Mon, 09 Jun 2025 18:02:39 +0000

Eight-second clips, decent visual richness, and Chinese text prompts that either get dropped or come out garbled.

Cursor Is Giving Away Three Months of Pro

Mon, 09 Jun 2025 14:00:52 +0000

A quick look at why the AI coding editor everyone’s been talking about is worth the free trial.

Starting Today, I'm a Google AI Pro Subscriber

Mon, 09 Jun 2025 14:00:36 +0000

Google I/O 2025 rebuilt every product around AI — enough to pull a committed OpenAI user over to a Pro subscription.

Clay-Style 3D Art, Made with OpenAI

Mon, 09 Jun 2025 14:00:24 +0000

Comparing Gemini’s and OpenAI’s image generation on the same blurry wall painting, then pushing the same subject into a 3D clay-render style.

Perplexity Pro, Free for a Year

Mon, 09 Jun 2025 14:00:10 +0000

Perplexity has joined the AI subsidy wars — a full year of Pro for free with a discount code, plus the usual advice to turn off auto-renewal.

Bored, So I Coded Up My Own Voice Clone

Sat, 10 May 2025 10:00:18 +0000

An open-source voice cloning project on a consumer GPU, then wired into my publishing pipeline so blog posts turn into short videos automatically.

Running Unlimited AI Translation Locally with Qwen3 4B

Fri, 09 May 2025 18:00:21 +0000

Pairing a small local Qwen3 4B model with Immersive Translate for a private, uncapped translation setup — plus a prompt trick that speeds it up.

Playing Around With ChatGPT-4o's New Photo Editing

Wed, 26 Mar 2025 23:01:20 +0000

Turning family photos into cartoons, adding a warship and a dinosaur to travel pictures, and restyling a portrait — a weekend with ChatGPT-4o’s new image tools.

Migrating My Image Host From a Self-Built Google Drive Setup to Chevereto

Sun, 24 Nov 2024 22:00:17 +0000

Swapping a 2TB Google Drive image host that needed re-authorizing every few days for a self-hosted Chevereto setup behind Cloudflare Tunnel.

Finally Subscribed to GitHub Copilot

Fri, 15 Nov 2024 11:03:26 +0000

Ten dollars a month for autocomplete that actually understands what I’m writing.

First Impressions of ChatGPT's o1-preview and o1-mini

Sat, 14 Sep 2024 07:17:01 +0000

A quiet rollout, a tight usage cap, and a version that’s clearly a big step up on reasoning.

My Free Image Host Died Overnight — and How I Rebuilt It

Sat, 07 Sep 2024 23:16:00 +0000

Telegram quietly killed Telegraph’s image-hosting support, breaking the free image pipeline my blog depended on — here’s how I patched it back together in a day.

Wrapping Whisper in a FastAPI Service for Remote Transcription

Sun, 25 Aug 2024 19:00:02 +0000

Offloading transcription from a GPU-less NAS to a desktop with an actual GPU, exposed as a simple internal API.

A Weekend with RAGFlow: Building My Own RAG-Based Knowledge Base

Sun, 25 Aug 2024 16:00:30 +0000

Deploying the open-source RAGFlow project, testing it against real documents, and wiring it up to local Ollama models to keep costs at zero.

Deploying Whisper on My NAS: Two Storage Gotchas

Sat, 10 Aug 2024 21:00:17 +0000

A simple script to auto-transcribe Obsidian voice notes, derailed twice by the NAS running out of space in root and /tmp.

Podsync Stopped Fetching New Episodes — Turns Out It Was Google

Sat, 10 Aug 2024 10:00:13 +0000

A morning-commute podcast habit broken by an upstream block, not anything on my end.

Trying Out Smart Connections: an AI Plugin for Obsidian

Thu, 08 Aug 2024 11:00:18 +0000

Chatting with my own notes via a local model, plus a gotcha when pointing the plugin at a self-hosted One API server.

Keeping Markdown Formatting Intact When Obsidian Syncs to Typecho

Sun, 04 Aug 2024 00:00:38 +0000

My Obsidian-based writing setup was publishing beautifully, except synced posts landed in Typecho full of raw HTML clutter — a hidden XML-RPC setting fixed it.

Fixing a Blank White Box Where Code Blocks Should Be in Typecho

Sat, 03 Aug 2024 23:00:20 +0000

A CSS/plugin conflict was rendering code blocks as an ugly blank space — turned out to be a syntax-highlighting plugin fighting the theme.

Comparing Open-Source LLMs for Customer-Service Conversation Summaries

Sat, 03 Aug 2024 12:00:04 +0000

Running ten locally-hosted models on consumer GPU hardware to see which ones best summarize customer-service calls after speech-to-text.

Fixing an Ollama Install Failure Caused by a Snap-Installed curl

Sat, 03 Aug 2024 00:00:31 +0000

Installing curl via snap put it ahead of /usr/bin on PATH, which broke the Ollama installer in a confusing way.

Reinstalling Ubuntu for a Dedicated LLM Box: NVIDIA Driver, CUDA, and cuDNN

Sat, 03 Aug 2024 00:00:30 +0000

After one too many driver crashes from installing random open-source software, I wiped the machine and set it up as a single-purpose LLM box.

AI Study Notes: Probability Density Functions and Weighted MSE

Wed, 31 Jul 2024 00:00:02 +0000

Working through two ML fundamentals in plain language: what a probability density function actually tells you, and how weighting by probability density changes mean squared error.

LLM Agents 101

Tue, 30 Jul 2024 00:00:09 +0000

What an ‘agent’ actually means in the context of large language models, and a worked example in a customer-service setting.

Installing an NVIDIA GPU Driver on Ubuntu 22.04

Sun, 28 Jul 2024 00:00:07 +0000

Notes on getting an NVIDIA driver and CUDA 12.4 installed cleanly, including disabling the nouveau driver first.

Fixing a CUDA Error When Running Gemma2 in Ollama

Sat, 27 Jul 2024 22:36:33 +0000

A one-line fix for a CUBLAS_STATUS_NOT_INITIALIZED crash: just update Ollama.

Fine-Tuning Your Own Model with Free GPU Compute

Fri, 10 May 2024 17:00:18 +0000

After AMD’s ROCm ecosystem let me down for local fine-tuning, free Colab GPUs turned out to be the pragmatic way to fine-tune Llama 3 for free.

Building a Personal Knowledge Base with MaxKB on My NAS

Fri, 10 May 2024 09:00:23 +0000

Self-hosting the open-source MaxKB knowledge-base tool at home, and pointing it at a Qwen-backed model for Chinese-language documents.

Running Llama 3 Locally with Open WebUI

Fri, 10 May 2024 09:00:11 +0000

Pairing a consumer GPU running local Llama 3 with Open WebUI and One API to combine local and remote models behind one interface.

Whisper + Obsidian: Fully Automatic Voice-to-Text Notes

Wed, 01 May 2024 13:00:11 +0000

Wiring up Obsidian’s Whisper plugin to the free Cloudflare-backed OpenAI API from the previous post.

Building a Free, OpenAI-Compatible API on Top of Cloudflare Workers AI

Wed, 01 May 2024 06:48:10 +0000

Wrapping Cloudflare’s free Workers AI models (including Qwen) behind an OpenAI-compatible endpoint, so existing front-ends don’t need to change a line of code.

Patching the Jasmine Theme to Show a Post's First Image as Its Thumbnail

Sun, 28 Apr 2024 17:05:00 +0000

Obsidian’s sync to Typecho doesn’t support custom fields over XML-RPC, so I patched the theme instead.

Two Quick Wins for This Blog: Caching and Visitor Analytics

Sun, 28 Apr 2024 13:57:08 +0000

Adding a caching plugin for faster page loads, plus wiring up Google Analytics.

International and Chinese Cloud Providers' LLM Landscape

Sat, 27 Apr 2024 18:00:00 +0000

A quick note on a comparison table of large-model offerings across major cloud providers.

Trying Out Local LLM Deployment: GPT-3.5 vs. Qwen-7B

Tue, 23 Apr 2024 15:30:00 +0000

Some hands-on experiments with un-tuned open models, and a few thoughts on where the real value in LLM products actually sits.

This AI Browser Extension Blew Me Away

Sun, 14 Apr 2024 18:09:00 +0000

Elmo can summarize any page on the fly — and it turns out that’s genuinely useful for reading in a second language.

Self-Hosting Immich on My Synology NAS: Batch-Importing My Photo Library

Sun, 14 Apr 2024 17:44:00 +0000

Getting Immich running on a home NAS and bulk-uploading years of photos with the CLI.

Google One Is Shutting Down Its VPN Service

Fri, 12 Apr 2024 10:25:07 +0000

Google says usage was too low to justify keeping the feature around.