Creative Coach

Type Work Project
Year 2026
Platform Slack Bot
Stack Node.js · Claude AI

Creative directors at Beatport review a high volume of work across campaigns, social, events, and editorial. A significant portion of that time gets spent on feedback that could have been caught earlier - wrong brand colours, unclear hierarchy, copy that doesn't match tone of voice. Designers wait for CD availability, the CD repeats the same notes, and the loop is slower than it needs to be.

Creative Coach is a Slack bot that auto-reviews creative work the moment it's shared in a channel. Drop an image (or paste a Canva link) into #creative-review and within seconds the bot responds with structured feedback - what's working, what needs fixing, and whether the work is ready for CD review. It's trained on Beatport's actual brand guidelines and gives feedback the way a senior creative would: specific, actionable, and referencing exact elements.

  • Drop an image or Canva link into Slack and receive structured feedback in seconds - no CD availability required
  • Verdict system - Strong / Needs work / Rethink needed - plus a clear "ready for CD review" flag
  • Two-section feedback format: what's working well and what can be improved, written with specificity (named elements, exact hex values, concrete actions)
  • "Deeper feedback" button expands per-dimension analysis across 8 evaluation criteria
  • "Add context and re-review" lets designers provide a brief, audience, and channel - the bot reruns the full review with that context factored in
  • Trained on Beatport's Brand Book 2025 and Voice, Tone and Style Guide 2025 - references specific guidelines when flagging issues
  • Persona and evaluation dimensions are fully configurable via persona.config.json - no code changes or redeployment required
  • New brands can be added by dropping PDFs into a folder and updating brands.json - zero code required
Node.js TypeScript Slack Bolt SDK Claude API SQLite Railway Claude Code GitHub
  • Before writing a line of code, three AI research agents were spawned in parallel - one investigating creative review workflows and pain points across agencies and in-house teams, one analysing UX patterns and AI tool positioning, and one synthesising findings into a full implementation plan
  • Research surfaced the insight that creative professionals reject tools that feel like a compliance checker - the "creative coach" framing and two-section feedback format came directly from this, not guesswork
  • The 8 evaluation dimensions (brand alignment, concept strength, message clarity, tone of voice, audience resonance, execution quality, visual hierarchy, contextual fit) were mapped from how creative directors actually evaluate work - not assumed
  • The is_blocking flag - distinguishing must-fix issues from nice-to-haves - came from research showing that treating all feedback as equal causes designers to feel overwhelmed and disengage
  • The two-section layout itself was a post-launch iteration: the original format was too long to read at a glance, and real usage triggered a redesign into a scannable surface view with deeper analysis behind a button
  • Built with Claude Code; deployed to Railway with auto-deploy from GitHub