-
-
Graph section
-
Novus
-
-
ContextCore
A personal learning and breakdown tool that takes a financial or tech headline — typed, pasted, or fetched from a URL - and traces its causal chain across finance, macroeconomics, and technology.
Rather than just summarizing news, it explains why something matters and how it ripples across domains, lets you keep asking follow-up questions in a conversational interface, and visualizes the underlying entities and relationships as an interactive knowledge graph.
Who it's for
Built primarily for myself. I'm transitioning from a CS background into fintech, and I wanted something that helps me genuinely understand the causal mechanics behind financial and tech news — e.g. how a Fed rate decision actually flows into tech stock valuations, or how an export ban ripples into trade tensions and semiconductor markets — rather than just chatting back and forth with a general-purpose chatbot and hoping the explanation sticks.
It's constrained to finance, tech, and macro topics by design, though it handles other text reasonably well too. That scope was a deliberate choice to keep the tool focused and the explanations genuinely useful rather than broad and shallow.
Tools used
- Python + Streamlit (UI, deployed on Streamlit Community Cloud)
- LlamaIndex — FunctionAgent for reliable native tool-calling
- OpenAI GPT-4o-mini
- Live Wikipedia API + NewsAPI integration
- streamlit-agraph for the interactive knowledge graph visualization
Architecture
A two-phase hybrid retrieval system:
- Live multi-source tool-calling — Wikipedia + NewsAPI gather grounding evidence per query, routed through a FunctionAgent
- LLM extraction — generates both a structured causal chain (step-by-step, domain-tagged) and a knowledge graph (nodes + edges) from that evidence
- Follow-up chat reasons over the already-built graph and research rather than re-querying live sources every time — graph-based retrieval for the conversational layer
What I learned shipping it
LLMs are tricky to work with :'(
Log in or sign up for Devpost to join the conversation.