Sonar — A&R Intelligence Agent

Sonar takes a natural-language brief from an A&R team and turns it into a fully-researched shortlist of artists, complete with scores, risk flags, and outreach drafts ready to send. It was built during an Enterprise AI Agent hackathon and won "Most Innovative Project."

The agent runs a multi-phase pipeline rather than a single prompt-and-response loop:

• Discovery — surfaces candidate artists matching the brief
• Cross-source validation — checks candidates against multiple data sources
• Deep per-artist research — builds a full profile per artist
• Competitive scan — situates each artist against comparable acts
• Scoring — produces a proprietary 0–100 "Sonar Score"
• Briefing — writes a full A&R brief with risk flags
• Outreach drafting — generates ready-to-send outreach emails

The orchestration layer runs on Anthropic Claude (Sonnet) via the Anthropic SDK, with prompt caching on the system prompt and tool schema to keep multi-turn runs fast and cheap, plus parallel/batched tool-calling for speed. Additional tool servers — Spotify, Reddit, Slack — plug in over the Model Context Protocol (MCP) without touching the core agent.

Redis Stack (via Docker) is the backbone for state and memory: session/chat history persistence, a RediSearch vector index (HNSW) for semantic artist-similarity search over Voyage AI embeddings, and Redis's Agent Memory Server for long-term qualitative memory and entity/topic tagging across sessions.

The frontend is built with Next.js (App Router), React, TypeScript, and Tailwind CSS, with a streaming chat UI that shows a real-time tool-call feed over server-sent events. External data comes from the Spotify Web API, YouTube Data API v3, Reddit API, and the Tavily API for web search and social-signal scraping.

Won "Most Innovative Project" at an Enterprise AI Agent hackathon.