... .. .. ... .. ...
x*8888x.:*8888: -"888: .=*8888x <"?88h. xH88"`~ .x8X
X 48888X `8888H 8888 X> '8888H> '8888 :8888 .f"8888Hf
X8x. 8888X 8888X !888> '88h. `8888 8888 :8888> X8L ^""`
X8888 X8888 88888 "*8%- '8888 '8888 "88> X8888 X888h
'*888!X8888> X8888 xH8> `888 '8888.xH888x. 88888 !88888.
`?8 `8888 X888X X888> X" :88*~ `*8888> 88888 %88888
-^ '888" X888 8888> ~" !"` "888> 88888 '> `8888>
dx '88~x. !88~ 8888> .H8888h. ?88 `8888L % ?888 !
.8888Xf.888x:! X888X.: :"^"88888h. '! `8888 `-*"" /
:""888":~"888" `888*" ^ "88888hx.+" "888. :"
"~' "~ "" ^"**"" `""***~"`
------------------------------------------------------------------------------
-- TABLE OF CONTENTS ---------------------------------------------------------
00. ABSTRACT .................................................. [ ]
01. WRITING ................................................. [ 02 entries ]
02. CODE .................................................... [ 10 entries ]
03. PODCASTS ................................................ [ 02 entries ]
04. AUDIO ................................................... [ 02 entries ]
05. CONTACT ................................................... [ ]
------------------------------------------------------------------------------
+============================================================================+
| 00. ABSTRACT |
+============================================================================+
matt — a multimodal generative model specializing in audio, video, and code generation. trained primarily on hands-on data since the late 2000s, with periodic offline fine-tuning. supports streaming inference, occasional batch processing, and one (1) concurrent user. context window degrades after roughly twenty hours awake. quantization not recommended.
------------------------------------------------------------------------------
------------------------------------------------------------------------------
+============================================================================+
| 01. WRITING |
+============================================================================+
feed: https://mattcavender.com/feed.xml format: RSS 2.0
------------------------------------------------------------------------------
01.01 On-Demand Spectrograms via GPU Compute ................ [ 2026-04-12 ]
How I replaced a pre-computed binary cache with on-demand WebGPU rendering for audio visualization.
01.02 I Vibe Coded a Color Grading Tool for Descript ........ [ 2026-04-12 ]
Descript's color grading UX wasn't cutting it, so I built my own — and ended up reverse-engineering their shaders.
+============================================================================+
| 02. CODE |
+============================================================================+
host: github.com/visionsofparadise
------------------------------------------------------------------------------
02.01 descript-color-grade ................................................
url: github.com/visionsofparadise/descript-color-grade
Color grading companion app for Descript. Runs Descript's actual extracted shader for pixel-identical preview across 7 adjustment sliders.
tags: [typescript] [shaders]
02.02 @e9g/spectral-display ...............................................
url: github.com/visionsofparadise/engineering/.../spectral-display
WebGPU audio spectrogram and waveform visualization. GPU-computed FFT, perceptual frequency scales, BS.1770-4 loudness metering, and true peak detection in a single streaming pass.
tags: [webgpu] [audio] [dsp]
02.03 @e9g/buffered-audio-nodes ...........................................
url: github.com/visionsofparadise/engineering/.../buffered-audio-nodes
Offline audio processing nodes for restoration, podcast production, and mastering. ML-powered transforms with ONNX runtime integration.
tags: [audio] [ml] [onnx]
02.04 @e9g/buffered-audio-nodes-core ......................................
url: github.com/visionsofparadise/engineering/.../core
Protocol and graph executor for composable audio processing pipelines. Streaming chunk-based architecture for arbitrary-length audio.
tags: [audio] [graph]
02.05 @e9g/buffered-audio-nodes-utils .....................................
url: github.com/visionsofparadise/engineering/.../utils
Shared DSP utilities for buffered audio processing — sample format conversion, resampling, and buffer management.
tags: [audio] [dsp]
02.06 DICES ...............................................................
url: github.com/visionsofparadise/dices
Quantum-resistant encryption layer using ML-KEM-1024 + X25519 hybrid with bounded triple ratchet. Provides forward/backward secrecy for P2P applications.
tags: [crypto] [p2p] [pqc]
02.07 DICE ................................................................
url: github.com/visionsofparadise/dice
P2P connectivity protocol using distributed peer infrastructure instead of centralized STUN/TURN servers.
tags: [p2p] [networking]
02.08 bufferfy ............................................................
url: github.com/visionsofparadise/bufferfy
Binary serialization library achieving 50-70% smaller output than JSON and 30-40% smaller than msgpack with full TypeScript type inference.
tags: [typescript] [serialization]
02.09 kademlia-table ......................................................
url: github.com/visionsofparadise/kademlia-table
XOR distance routing table for Kademlia DHT with extendable k-buckets, LRU ordering, and external liveness management.
tags: [p2p] [dht]
02.10 sqs-mock ............................................................
url: github.com/visionsofparadise/sqs-mock
A functional, in-memory mock of the SQS client from AWS-SDK.
tags: [aws] [testing]
+============================================================================+
| 03. PODCASTS |
+============================================================================+
------------------------------------------------------------------------------
03.01 Pretrained ..........................................................
url: pretrained.fm
Weekly AI/ML podcast hosted by Pierce Freeman and Richard Diehl Martinez covering LLM development, infrastructure, and industry trends.
tags: [ai] [ml]
03.02 Working Code ........................................................
url: workingcode.dev
Weekly developer culture podcast. Four hosts discussing career growth, soft skills, and tech industry culture (not code tutorials).
tags: [culture] [career]
+============================================================================+
| 04. AUDIO |
+============================================================================+
------------------------------------------------------------------------------
04.01 sonny banks .........................................................
url: open.spotify.com/artist/6ib0DdCLBUPWNaHztQhrSL
bass house music (est. 2014).
tags: [bass-house]
04.02 xkore ...............................................................
url: open.spotify.com/artist/5cfALf5vLhWIo7VeIh4fQX
dubstep and drum & bass music (est. 2008).
tags: [dubstep] [dnb]
+----------------------------------------------------------------------------+
| |
| [@] greetz@mattcavender.com |
| |
+----------------------------------------------------------------------------+
==============================================================================
--- EOF --- 2026-04-25 --- 0x4D43 ---