I shipped the first version in about two weeks, validated it with a handful of content strategists through shadowing sessions, and iterated on their workflows to reduce manual research time and make their daily brief preparation more reliable.

• Designed a multi-stage LLM pipeline that combines web search with RAG over client transcripts to filter out noisy results and surface only relevant, on-brand news and talking points for upcoming content.
• Used a couple of focused n8n workflows to handle integrations and scheduled runs, wiring the system into existing tools for content strategists, while keeping most of the RAG and LLM refinement logic inside the Next.js app.
• Extended an existing Next.js + Supabase internal tool with a shadcn-based UI to configure clients, inspect runs, and debug issues, and introduced Langfuse to monitor prompts, latency, and failures across the pipeline.

The tech stack includes Next.js, TypeScript, Supabase, OpenRouter, Langfuse, n8n, and shadcn/ui.