Ctrl + Shift + A to toggle
Personal ventures and notable engineering projects — from founding a product to deploying production infrastructure at scale.
Products and platforms I've built independently.
Distributed video transcoding for on-prem hardware — slices any video into keyframe-aligned chunks, dispatches them across every encoder on the LAN, and stitches the output back together. Built as solo technical founder from chunking algorithm through frozen exes.
Owns the job queue, the SQLite state store, the web UI, and every chunk's signed HTTP transfer. One brain per fleet.
Probes hardware, registers via mDNS, pulls chunks over signed URLs, and streams live progress back over a bidi gRPC channel.
Reads the source's keyframe index via ffprobe and picks GOP-aligned boundaries — no source re-encoding for closed-GOP H.264 / HEVC.
Scores idle workers against pending chunks on encode speed (EMA), network distance, queue delay, and capability fit — lowest cost wins.
YouTube 4K HEVC, Instagram Reels, Premiere Pro proxies, ProRes 422 HQ, DNxHR HQX, AV1 — every flag pinned, every output verified.
PyInstaller-packaged master.exe and worker.exe — Python runtime, ffmpeg, and ffprobe baked in. No installer, no PATH wrangling.
Local-first AI desktop app for photographers, agencies, and stock teams — tags thousands of images with a vision-language model running on the user's own machine, then develops them in a Lightroom-style editor. Every byte round-trips with Adobe Bridge, Lightroom Classic, and NeoFinder. Built as solo technical founder.
Walks a folder, runs every supported image through a local vision-language model, and writes natural-language captions + 12–20 keywords into standard XMP / IPTC fields.
Exposure, tone curve, split toning, vignette, grain, crop, AI background removal, 2×/4× upscale, LaMa generative fill. Every slider writes to Adobe Camera Raw's crs:* XMP namespace.
Preview folders before importing — no DB writes, fast thumbnail grid, metadata-aware filter chips, rate / pick / reject keyboard shortcuts.
Opt-in InsightFace buffalo_l + sqlite-vec kNN — drop a reference photo, find every other photo of that person across workspaces. EULA-gated.
Browse, pull, and benchmark vision models from inside the app. Knows which models fit your VRAM and which licenses require explicit acceptance.
CSV, JSON, Lightroom XMP sidecars, NeoFinder XML, Markdown manifests, hierarchical Keywords.txt — plus AES-256-GCM encrypted .gvdb archives for Studio.
Browser-based professional audio platform — six studio-grade tools and a full audio engineering education system, built as solo technical founder from architecture through product.
A node-based visual processor where users build custom signal chains by connecting effects — compressors, limiters, noise gates, parametric EQ, filters, distortion, modulation, spatial effects, pitch/time tools, and mastering processors. Built on React Flow with a real-time audio engine.
State-of-the-art neural models separate any track into vocals, drums, bass, and instruments. Supports standard presets (vocals only, instrumental, 4-stem, 6-stem) and an advanced mode. Accepts MP3, WAV, FLAC, AAC, OGG, and Opus up to 100MB.
Intelligent mastering that matches a track to a professional reference — either a genre preset (Rock, Pop, Hip-Hop, Classical, Electronic) or a user-uploaded reference. Outputs WAV, MP3, or FLAC. Handles loudness targeting, dynamics, and spectral balance automatically.
Converts any audio recording into a playable MIDI file using AI pitch detection. Instrument-optimized presets for vocals, piano, guitar, bass, and mixed recordings. Supports batch processing of up to 5 files simultaneously and live microphone recording up to 15 minutes.
Preset-driven noise removal and voice processing covering podcast cleanup, heavy noise reduction, voice clarity, broadcast voice, and interview cleanup. Also includes music mastering and normalization categories. Processes any format up to 100MB.
21 modules, 263+ lessons, 205+ hours of content covering everything from sound physics to advanced DSP and genre-specific production. 13 core modules (fundamentals through EDM production) plus 8 genre specializations (Rock, Hip-Hop, Metal, Country, R&B, Post-Production, Jazz, Podcast). Includes 600+ quiz questions, XP progression, ear training, mix challenges, and 21 certificates. First lesson of every module is free.
Cloud rendering platform built on AWS Deadline Cloud — modern GPU infrastructure, DCC-native submitters, and transparent per-node-hour pricing for VFX professionals.
Significant engineering deployments and infrastructure builds at GPL Technologies.
Proof-of-concept engagement validating AWS Deadline Cloud for a full CBFX production pipeline — including cross-account RLM license connectivity and a hardened Rocky Linux 9 artist workstation image.
Production deployment of the full CBFX render farm on AWS Deadline Cloud — from infrastructure-as-code foundation through a Rocky Linux 9 artist workstation image used studio-wide.
Custom Service Managed Fleet worker build that brought licensed NeatVideo OFX plugins to Deadline Cloud — a solution AWS did not ship out of the box.
Production-ready CloudFormation template deploying a full Houdini 21 render farm on AWS Deadline Cloud — farm, queue, fleet, S3 attachments, IAM, and monitoring in a single stack.
POC proving out GPU rendering on AWS Deadline Cloud with two parallel queues — Cinema 4D 2026 with Redshift 2026.1, and Cinema 4D 2026 with C4DtoA (Arnold) — using Service Managed Fleets and Spot capacity.
Multi-phase AWS-funded engagement standing up a startup VFX studio on AWS — from network architecture through cost modeling, security, and delivery timeline — in partnership with AWS's Startup VFX program.
Proof-of-concept engagement moving JAMM VFX workloads onto AWS Deadline Cloud with a studio-specific SMF worker configuration and queue topology.
Proof-of-concept deployment validating Deadline Cloud as a render backend for Buck's motion design and VFX pipeline.
Proof-of-concept for Cantina Creative that had to beat on-prem render hardware on both speed and cost — benchmarked against a real 20K-resolution, 2,000-frame production sequence.
Proof-of-concept engagement deploying AWS Deadline Cloud infrastructure and validating end-to-end render workflows for Sight & Sound.
Production deployment of a custom Maya tile-render solution on AWS Deadline Cloud, enabling distributed tile-based rendering of oversized Maya frames across the SMF worker fleet.