INSPECTORR
Production-grade sentiment analysis for finance
Real-time processing of analyst notes and market commentary built with Next.js, FastAPI, Python and Claude AI (Anthropic)
Click above to see the AI sentiment analysis pipeline in action
What this does
- •Daily per-asset insights: summary, key drivers, risk cues, tone, confidence.
- •Portfolio view: quick read of where sentiment is rising, turning, or diverging from price.
- •Auditability: each insight ties back to dated sources and an explicit prompt version.
- •Professional clarity: clean language that reads like an internal note.
How it works
- •Generates sample data: analyst notes, investor feedback, market commentary.
- •Ingest & clean: normalizes data into asset_id, text, source_date, sentiment_score with basic hygiene.
- •Model & summarize: Claude reads per asset/day and returns summary, drivers, risks, tone, confidence (versioned prompt).
- •Serve & visualize: A FastAPI backend exposes insights and sentiment aggregates for the next.js UI.
Scale Path
- ▸More sources: broker research, transcripts, earnings snippets, social finance
- ▸Model flexibility: additional providers; route by cost, latency, or sensitivity
- ▸Chunking & routing: smart grouping per asset/day with retries, backoff, budget guards
- ▸Caching & history: snapshot insights by day for trend lines and “what changed” diffs
- ▸Role-aware views: PM overview, analyst deep-dives, compliance-ready exports
Key Product Principles
- ✓Signal over noise: concise, explainable outputs that can be scanned in seconds
- ✓Governable by default: versioning, clear provenance (method, confidence, source dates)
- ✓Cost-aware: batching, token caps; easy to swap models or set demo limits
- ✓Separation of concerns: pipeline, backend, frontend clearly separated
- ✓Enterprise-ready: SSO and workspace governance