Selected work

Production AI systems that shipped.

Case studies in AI cost optimization and video production automation — each one with the numbers.

1.3M+ views and 2,500+ followers in five days

An autonomous system that generates on-voice posts with realistic visuals, publishes to Threads on one-tap approval, and learns what converts. Off a standing-still account.

Python · FastAPI · Celery · pgvector · LLM ensembleRead case study →

Video automation

~3 min

Automated processing (was ~1 hr manual)

~1 hour of manual editing per video → ~3 minutes automated

A managed pipeline for a London video agency: raw lesson recordings in, finished brand-consistent videos out. AI cleanup, pixel-accurate reframing, human approval on cuts.

Python · FFmpeg · OpenCV · Whisper · GeminiRead case study →

Cost reduction

99%+

Cost reduction vs proprietary API

99%+ cost reduction vs premium video AI

Replaced a premium proprietary video AI service at $3–5 per minute with a custom ComfyUI workflow built on Infinity Talk and Wan 2.1. Per-video costs dropped from dollars to cents.

−99%+ cost

ComfyUI · Infinity Talk · Wan 2.1 · DockerRead case study →

Cost reduction

84%

Cost reduction at production scale

84% cost reduction with capability premium services can't match

Replaced premium proprietary motion-control video services with a custom ComfyUI workflow on Wan 2.2. Approximately $12,000 in annual savings at the client's production scale.

−84% cost

ComfyUI · Wan 2.2 · RunningHub · FFmpegRead case study →

Multi-model pipeline

< $1

Per 20-minute video produced

Multi-model AI pipeline for video localization at scale

Architected a complex multi-model AI pipeline combining semantic embedding generation, self-hosted speech transcription, LLM-based script rewriting, and multi-language voice synthesis.

Vertex AI · Whisper · Gemini · QdrantRead case study →

Production SaaS

3 months

Solo development to launch

Production SaaS for Telegram content automation

Solo-built end-to-end SaaS platform automating content creation for Telegram channels. Architected multi-agent LLM orchestration with cross-verification that prevents typical AI failures.

FastAPI · React · PostgreSQL · Multi-agent LLMRead case study →