Skip to content

ADR 0007: AI Processing — Tencent CI + MetaInsight

Status

Accepted

Date

2026-07-24

Context

OVES generates and processes significant volumes of images, videos, documents, and unstructured data. AI processing is needed for:

  • Product image transformation (resize, crop, watermark)
  • Video processing (transcode, snapshot, smart cover generation)
  • Document conversion (Office → PDF, preview generation)
  • Content moderation (images, videos, audio, text)
  • Semantic search across document archives
  • Speech recognition and synthesis

The existing media processing pipeline relied on Cloudinary for transformations and lacked a unified AI processing layer. With the move to Tencent COS for all storage, there is an opportunity to use COS-native AI services to eliminate data movement costs.

Decision

Tencent CI (数据万象) and MetaInsight are the primary AI processing services. All processing is COS-native — no data egress required.

Capability Mapping

Capability Service Integration
Image resizing, cropping, formatting, watermarking CI COS → CI → result back to COS
OCR text recognition CI Inline via COS API
Video transcoding, snapshots, smart cover CI COS trigger → async job
Office document → PDF CI COS trigger → async job
Document preview (image/HTML) CI On-demand via COS URL
Content moderation (image/video/audio/text) CI Sync or async audit
Speech-to-text, text-to-speech CI Async job via COS API
Noise reduction, voice separation CI Async job via COS API
Image search (search by image/text) MetaInsight Dataset + binding to COS bucket
Hybrid document search MetaInsight DocSearch dataset on COS bucket
Knowledge base (one-click) MetaInsight Auto-creates bucket + dataset + binding

Processing Pipeline

Upload → COS → CI/MetaInsight (sync or async) → write result back to COS → application reads

Tooling

  • WorkBuddy skill: tencentcloud-cos (installed) provides unified CLI access via scripts/cos_node.mjs
  • Programmatic: Tencent COS SDK + CI API for custom pipelines

Consequences

Benefits

  • Zero data movement: processing happens within the same COS bucket; no egress costs
  • Unified API: COS + CI + MetaInsight accessed through a single skill/script
  • One-click knowledge base creation (COS bucket + DocSearch dataset + binding)
  • Chinese-language content processing optimized (OCR, speech recognition, content moderation)
  • No separate AI service subscription needed — CI and MetaInsight are pay-per-use on top of COS

Trade-Offs

  • Tencent CI/MetaInsight are Tencent-specific; not portable to other cloud providers
  • Cloudinary remains for global CDN distribution; dual processing layers (Cloudinary transforms for delivery, CI for AI processing) adds configuration
  • MetaInsight knowledge base feature requires specific dataset template configuration (DocSearch)

Supersedes

None.

Superseded By

None.