PulseLens - Evidence Market Intelligence

PulseLens is an evidence-backed market intelligence workspace for fast-moving sectors. The current demo tracks the US AI Hardware / Semiconductor market - Nvidia, AMD, and Supermicro. Market signals are scattered across SEC filings, investor relations pages, pricing pages, and news. Analysts spend too much time collecting and cross-checking sources, and generic AI summaries offer no way to audit where each claim came from. PulseLens addresses this with a 14-node multi-agent pipeline grounded in 11 peer-reviewed methods: Step-Back Prompting and Multi-HyDE (EMNLP 2025) for query diversity, RASG for schema-constrained fact extraction, SAFE (Google DeepMind 2024) for atomic claim verification, FinBERT for financial sentiment scoring, ClaimCheck (ACL 2025) and FActScore (EMNLP 2023) for cross-source triangulation, STORM (NAACL 2024) for narrative synthesis, and Self-RAG (NeurIPS 2023) and FLARE (EMNLP 2023) for the grounded chat graph. A strict quality gate enforces minimum fact count, source diversity, and signal coverage before a report is accepted. Financial analysts, market researchers, competitive intelligence teams, and sector-focused investors can use PulseLens to inspect source-backed evidence with full quote traceability, explore GPU pricing pressure, compare company momentum and narratives, analyze signal depth across seven signal types, review the full pipeline audit trail, and ask a grounded chat assistant questions with any fact, signal, company, or risk alert attached as context - every answer cites the exact source it came from. Built on Bright Data for live web collection, LangGraph for agent orchestration, FastAPI, OpenRouter, and Supabase/Postgres, with a React/Vite/TypeScript frontend. Research demo only - not intended for investment or trading decisions.