01 / GENERATIVE AI & VFX PRODUCTION

Generative AI & VFX Production

Hybrid CG/ML pipelines built by people with twenty-plus years of feature-film credits. The tools are new; the shot discipline is not.

AI Image & Video Generation Pipelines

Production pipelines that take generative AI from experiment to deliverable. We design ComfyUI graphs and multi-model workflows for storyboards, concept art, previs, and final pixels, with LoRA training and cloud video models where they earn their place. IntelliStory, our structured generative production platform, runs this entire stack from first idea to final render.

  • ComfyUI workflows
  • Multi-model orchestration
  • LoRA training
  • Runway / Kling / Veo

Hybrid CG/ML VFX Tooling

State-of-the-art models packaged as artist-first tools. Our custom Python deploys bundle Netflix Void cleanplates, Corridor Key chroma-free keying, SAM3 and MatAnyone video mattes, Depth Anything, GVHMR body-mesh recovery, and DWPose skeletons behind drag-and-drop UIs. Artists get shot-ready mattes, cleanplates, depth passes, skeletons, and scene-linear 360° HDRIs for Maya, Houdini, Blender, and Nuke without ever opening a node graph.

  • Custom Python deploys
  • Video mattes & cleanplates
  • Depth / normals / pose
  • Plate → 360° HDRI

3D Reconstruction & Gaussian Splatting

Single images turned into navigable 3D. We build Gaussian splatting pipelines and production-ready WebGL viewers, from Apple SHARP inference generating over 1.1 million splats in under a second to React Three Fiber asset libraries with live in-browser previews.

  • Gaussian splatting
  • Apple SHARP
  • WebGL / React Three Fiber
  • Instant 3D reconstruction
Proof Sharp · 3D Vault

VFX Supervision & Art Direction

Twenty-plus years of feature-film VFX behind every engagement: environment TD on The Lord of the Rings: The Return of the King at Weta, credits on Frozen, CG Supervisor on Dune: Part Two at DNEG, and Art Director on TCL's inaugural FILM MACHINE program. We supervise traditional shows and AI-augmented productions alike, and we know exactly where generative tools do and do not hold up on a real shot.

  • CG supervision
  • Compositing supervision
  • Art direction
  • AI-augmented production
02 / AGENTIC SYSTEMS & LLM ENGINEERING

Agentic Systems & LLM Engineering

Agents that inspect their own output and iterate until it meets the brief, plus the integration and context engineering that makes LLMs reliable in production.

Agentic AI & MCP Development

Agents that close the loop instead of stopping at "no errors." We build MCP servers and agentic workflows where the model runs the tool, looks at the actual output, critiques it, and iterates until the result meets the brief. Our open-source ComfyUI Loop MCP is the pattern in public: build, run, look, critique, fix. Hello World takes it further, an app that LLM agents rewrite from user tickets.

  • MCP servers
  • Build → run → look → fix loops
  • Claude / GPT tool use
  • Self-evolving apps

LLM Integration & Intent Parsing

Natural language in, structured action out. Our systems convert plain-English intent into structured task primitives: commander intent into mission plans for a defense tactical simulator, art direction into procedural tile sheets, health questions into coaching grounded in the user's actual labs and vitals rather than generic advice.

  • Intent parsing
  • Structured outputs
  • Multi-LLM orchestration
  • Grounded AI assistants

Context Engineering & Notation Systems

MOTH, our machine-optimized text hierarchy, compresses specs and PRDs by 70 to 90 percent while staying readable. kablUI describes interfaces in 5.5 percent of the characters of equivalent HTML. We apply the same discipline to your prompts, docs, and agent context windows.

  • MOTH notation
  • kablUI
  • 70-90% compression
  • Prompt engineering
Proof MOTH · kablUI · Promptobot
03 / DATA & INFRASTRUCTURE

Data & Infrastructure

The unglamorous layer that decides whether AI works in production: GPUs, queues, datasets, provenance, and storage that survives an audit.

GPU Infrastructure & Model Serving

The orchestration layer under our own AI stack, available for yours. We run a queue-driven, multi-tenant GPU platform on an 8× NVIDIA B200 host: Redis and BullMQ job queues, Docker Compose services on a shared CUDA + PyTorch base image, strict per-tenant GPU isolation, and access brokered through a Tailscale jump host so the cluster never touches the public internet.

  • NVIDIA B200 / CUDA
  • Redis + BullMQ queues
  • Docker Compose
  • Tailscale-secured

Training Data & Dataset Engineering

Datasets built for the model you actually want. Our curation pipelines span video frame archives, manifest generation, and VPR satellite imagery, with quality-control gates that hold at scale instead of degrading quietly as the dataset grows.

  • Frame archives
  • Manifest generation
  • Satellite tiles
  • Large-scale curation

AI Metadata, Provenance & Content Analysis

Detection and verification for AI-generated content. We build C2PA and IPTC provenance checks, binary pattern analysis, and ffmpeg-based batch pipelines that tell you what a file really is and where it came from. MetaProbe, our media metadata extractor, is the working demonstration.

  • C2PA & IPTC detection
  • ffmpeg batch processing
  • Content authenticity
  • Binary pattern analysis
Proof MetaProbe

Digital Asset Management

Enterprise DAM with immutable versioning and VFX shot tracking. Our reserve-upload-commit pattern keeps every version durable, background workers keep ingest fast, and CAD/PDM integration extends the same discipline to aerospace and manufacturing.

  • Immutable versioning
  • Reserve → Upload → Commit
  • Background workers
  • CAD/PDM integration
Proof 3D Vault

Cloud Architecture, Storage & DR

AWS architecture (EKS, RDS, S3) for enterprise scale and Supabase + S3 patterns for rapid deployment. Durable job queues, DR runbooks, SLO monitoring, and observability come standard.

  • AWS EKS/RDS/S3
  • Supabase storage
  • DR & observability
  • SLO monitoring
04 / PRODUCT & WEB ENGINEERING

Product & Web Engineering

Complete, live products rather than prototypes. Most of the apps linked below are running right now, and you can try them before you hire us.

Full-Stack Product Development

React frontends, typed APIs, Postgres and Supabase backends, auth, realtime, and PWA installs. Live examples include a personal-health platform with an AI coach grounded in your own labs, real-time classroom polling that students join with a 4-letter code, and an AI type-pairing studio built on the full Google Fonts library.

  • React / Vite
  • Supabase + Postgres
  • PWAs
  • Realtime apps

In-Browser Media Tools & WebAssembly

Heavy media processing with zero upload. Using ffmpeg.wasm and on-device model inference we ship tools that matte people out of video, mux audio into clips, restyle footage as ASCII or pop art at up to 4K, and produce print-ready PDFs, all entirely client-side. Privacy review gets simple when no frame ever leaves the machine.

  • ffmpeg.wasm
  • Client-side AI inference
  • Zero-upload privacy
  • Up-to-4K export

AI-First Rapid Prototyping

Our own catalog is the argument: dozens of live tools shipped by a small team working agent-first with Claude Code and custom in-house tooling. We bring the same workflow to your team, build the first prototypes with you in days, and teach your engineers to run the loop themselves.

  • Claude Code / agentic dev
  • Days, not quarters
  • Live deployments
  • Team enablement
HOW WE ENGAGE

Four Ways to Work With Us

/AUDIT

Pipeline Audit

A focused read of your current pipeline, model stack, or infrastructure. You get a prioritized roadmap of what to fix, what to build, and what to leave alone.

/BUILD

Build & Handoff

We design, build, deploy, and document a tool or pipeline, then hand it over with your team trained to own it. You keep it running without us.

/EMBED

Embedded Development

Senior engineering inside your team on a fractional or full-time basis, whether that means supervising an AI-augmented show or standing up your GPU platform.

/TRAIN

Training & Enablement

Hands-on workshops in ComfyUI, agentic development, and LLM integration, so your artists and engineers ship with these tools instead of watching demos.

Have a pipeline problem or a product idea?

Tell us what you are trying to ship. We will tell you honestly whether AI helps, what it costs, and what we would build first.

Get in Touch See the Work