Project examples

Problem

Teams lose time searching across documents, shared drives, notes, and internal tools to find information they need to do routine work.

Approach

Structure internal knowledge around real operating workflows and make it accessible through natural-language retrieval.

System

Document-aware assistants, internal search, permission-aware retrieval, and workflow integrations built around company knowledge.

Outcome

Fewer repeated questions, faster internal decisions, and less time spent hunting for answers.

Problem

Some companies need AI capabilities but cannot send sensitive data, internal records, or operating context to third-party cloud providers.

Approach

Design local-first systems that preserve privacy and control while still delivering practical AI workflows.

System

On-premise inference, private document handling, local runtimes, and secure internal workflows tailored to operational constraints.

Outcome

Teams can adopt AI in real work without compromising privacy, compliance, or control over sensitive information.

Problem

Teams producing serialized video, especially in multiple languages, repeat the same operational steps every release: scripting, voice work, timed assembly, and platform publishing. Manual handoffs between each step limit volume and consistency.

Approach

Design a single pipeline where script generation, voice synthesis, visual assembly, and distribution run as connected operational stages with clear inputs, outputs, and quality checks.

System

A working end-to-end stack: LLM-assisted script generation, ElevenLabs voice cloning across languages, Whisper-synced animated assembly in Remotion, and automated YouTube publishing. Originally built for a language-learning series; now extended to support content production for The 60 Second Film Festival, an international one-minute showcase running since 2017, with 300+ submissions per cycle and sponsors including YouTube and Epson.

Outcome

Production moves from script to published episode with minimal per-release manual work, reliable multi-language output, and infrastructure that supports ongoing series volume without adding proportional editing headcount.