MarketingSaaS · Web application · AI/ML · Integration · Automation · Internal platform
AI-powered conversion-optimisation agency platform
Extended an e-commerce optimisation agency's internal AI platform with real-time test-data sync, prioritisation and sprint planning, automated client slide decks, LLM research agents and embedding-based retrieval groundwork, and cut the LLM cost of its review-mining pipeline.
Problem
The agency's A/B test data, client-call insights and reporting were spread across external tools, and it needed reliable pipelines, planning features and AI-assisted research on top of that data, at controlled LLM cost.
Solution
The team fixed production issues and then built a webhook-driven real-time sync of test tasks into Supabase (field mapping, screenshots, status history, signature checks, idempotency, deletion sync, write-back of generated test IDs), prioritisation scoring and a sprint planner, a service-account slide-deck builder, comment embeddings in pgvector for future retrieval, transcript persistence, an LLM brand-voice research agent split across chained calls to fit serverless limits, a rich-text editor migration, and review-mining optimisations that cut LLM calls by about 40% and database write load by about 75%.
Key features
- Real-time webhook sync of A/B test tasks with signature verification and idempotency
- One-time mass import of 853 test tasks
- Write-back of auto-generated sequential test IDs and app links via triggers and edge functions
- Prioritisation scoring and per-client sprint planner
- Automated slide-deck builder from templates with stage-aware placeholders
- Comment and thread embeddings in pgvector across 15 client workspaces
- Meeting-transcript persistence and test-idea extraction with managed prompts
- LLM brand-voice research and validation agents with web scraping
- Rich-text editor migration with sanitisation and markdown conversion
- Review-mining pipeline with checkpointing and crash recovery
Tech stack
- Anthropic
- Claude
- ClickUp API
- DOMPurify
- Firecrawl
- Fireflies.ai
- GA4 API
- GitHub
- Google Slides API
- Google service accounts
- Langfuse
- Next.js
- OpenAI
- PostgreSQL
- React
- React Query
- Supabase
- Supabase Edge Functions
- Supabase Storage
- TipTap
- Tuleap
- TypeScript
- Vercel
- Weaviate
- n8n
- pgvector
- text-embedding-3-small
Services behind this project
Read more
22SEPRAG vs fine tuning: how to add an LLM to your product
RAG vs fine tuning: which should you use to add an LLM to your product? A plain guide to choosing, and to the production work each approach really needs.
AI & Automation6 min read
18AUGHow we staff an engagement: engagement lead, project manager, technical lead
The people on a project matter as much as the stack. Here's how we size a team from a shared bench, with a named engagement lead, project manager and technical lead on every engagement.
Process3 min read
04AUGShipping AI features into real products
Patterns that repeat across the AI products we've built: transcription pipelines, retrieval over your own data, a human kept in the loop, and controls on what the model costs and can see.
AI & Automation3 min read