Cognitive / Investment Intelligence
AI VC - Investment Intelligence
Solo-built investment intelligence service that ingests news/trends/macro indicators, runs a theme workflow (LangGraph when available, sequential fallback otherwise), and provides RAG-based Q&A with PII redaction and multi-provider routing/fallback.
- Role
- Solo builder · architecture, implementation, and operational tooling
- Period
- 2025 - Present
- Context
- Personal Project
Solo-built investment intelligence service that ingests news/trends/macro indicators, runs a theme workflow (LangGraph when available, sequential fallback otherwise), and provides RAG-based Q&A with PII redaction and multi-provider routing/fallback.
What made the system difficult.
Needed reliable scheduled ingestion with backoff/dup handling, plus orchestration that can route across multiple LLM providers and keep theme state consistent over time.
How boundaries and flows answered it.
Implemented ETL pipelines with PostgreSQL upserts, a multi-provider router with fallback policies, and a StateGraph orchestrator for theme analysis. Added a RAG pipeline with PII masking + retrieval to support evidence-based theme queries, driven by APScheduler jobs.
Overview
- Solo-built investment intelligence system focused on theme discovery, tracking, and Q&A.
Purpose & User Flow
Purpose
- Detect and track investable themes over time and answer questions with evidence over ingested sources.
User Flow (high level)
- Scheduled ETL (news/trends/macro) → theme normalization/modeling → report outputs → RAG Q&A with evidence (PII-safe).
Key Components
- ETL pipelines for GDELT, Google Trends, FRED with PostgreSQL upserts
- Multi LLM provider router + fallback policies
- LangGraph(StateGraph) orchestrator for theme lifecycle (collect → normalize → model → report)
- RAG pipeline with PII redaction and vector retrieval
- APScheduler runner for scheduled ingestion and analysis execution