SnitchFeed?

Create an original replacement for SnitchFeed's intent-based social listening workflow, but do not copy its brand, logo, copy, or UI. The core workflow: a user creates a 'listener' by describing in plain text which keywords, competitor brands, intent signals (e.g., buying intent, pain point, competitor complaint), and exclusions matter. The backend continuously polls/streams public posts/comments from Reddit, Bluesky, X, and LinkedIn. Each collected item is sent to an LLM that returns a fit score (0-100), sentiment, intent labels, and a short justification. Only items passing user-set score/intent thresholds are delivered as alerts to Slack, Discord, email, or a webhook endpoint, which can forward to a CRM or Zapier. Requirements: user-facing web app with listener CRUD, live feed of scored mentions, threshold controls, alert channel config, and webhook log; backend with queue-based workers, API clients for each platform (use official APIs where available, avoid scraping if ToS-problematic), SQLite or Postgres persistence, and an LLM provider (OpenAI-compatible API) for classification. Implementation approaches: Next.js + TypeScript for frontend and API routes, Redis or BullMQ for queueing, Python FastAPI or Node workers for ingestion/classification, webhook delivery with retries. Deliberate out-of-scope: billing, multi-tenancy auth beyond single admin login, mobile apps, LinkedIn auth-scoped scraping, and any trademarked branding. Include a README with setup instructions, platform API/rate-limit notes, LLM key configuration, webhook security caveats, and known limitations (e.g., platform ToS, delayed polling on some networks, LLM misclassification risk).