Skip to main content
JinzaiLab Ching Hsu's AI automation studio

Selected work

Projects and practical outcomes

Work across media, documents, research, and cloud integrations. Each summary focuses on the problem, the system, and measurable results.

Client data and sensitive implementation details have been removed. Figures are based on existing project records.

Media workflow automation

AI video editing and asset delivery tool

Combined Faster-Whisper transcription, LLM clip analysis, FFmpeg editing, metadata generation, and FTPS delivery in a cross-platform desktop tool.

Reduced editing time by about 80% based on existing production records.

PythonFaster-WhisperOpenAIFFmpegDesktop

Video content analysis

Hybrid video metadata system

Combined Google Video Intelligence and Gemini to turn detected objects, scenes, and production techniques into stock-platform-ready English descriptions and keywords.

Supports 18 categories, with recorded cost estimates of US$0.0075–0.225 per video.

Video IntelligenceGeminiMetadataCost analysis

Generative media

Multi-provider AI video generation tool

A single interface for Vidu and Google Veo with CSV input, Google Drive, batch jobs, progress tracking, pause controls, and retry handling.

A modular codebase of more than 2,500 lines reduces the work required to switch generation providers.

ViduGoogle VeoBatch processingPython

Document intelligence

AI pitch deck evaluation system

Parsed PDF and PPTX decks for structured evaluation across team, technology, R&D, business, and presentation quality.

Evolved through four major versions into a product specification with accounts, history, and billing.

Multimodal AIPDFPPTXGoogle Docs API

Document intelligence

Multi-format application document checker

Handled more than ten formats—including PDF, OCR, DOCX, PPTX, and ZIP—to review required files, completeness, and risk.

Organized document inventory, overview scoring, and risk analysis into one consistent three-layer review.

OCRDocument AIRisk analysisPython

Workflow automation

Company data validation workflow

Connected CSV data, structured OpenAI validation, exception retries, and HTML email reports in n8n.

Nineteen nodes cover input, AI review, error handling, and notifications.

n8nOpenAICSVEmail

Research automation

AI grant information search

Collected public grant pages, extracted eligibility and deadlines, and turned the results into a readable research response.

Recorded test cost was about US$0.01 per query.

Web scrapingOpenAIBeautifulSoupResearch

Research and content

Research-to-content agent pipeline

A four-stage workflow for research QA, content strategy, channel adaptation, and consistency review, with human checkpoints.

Preserves claim traceability and source requirements across every stage.

AI agentsContent strategyQATraceability

Your workflow may look different

Tell me about the tools, data formats, and steps taking the most time. I can help identify a focused place to begin.

Tell me about your project