Source-grounded stock chat
The chatbot securely calls external financial tools and only answers with grounded evidence from reliable sources.
Independent project: an AI system on finance
YoungStocks turns live market data into grounded AI financial analysis.
A production-ready web platform that demonstrates tool-calling orchestration, multi-agent reflection, and RAG-backed domain expertise in one source-grounded experience.
- 3AI interaction models
- Liveweb app at YoungStocks.org
- Groundedanswers backed by real sources
Core capabilities
Three AI interaction models working as one product.
Debate before recommendation
A trading agent proposes a plan while other agents challenge reasoning, constraints, and risk assumptions to reduce hallucinations.
Reusable domain context
The RAG system acts as both an expert assistant and a context service that can feed other system agents with higher-fidelity financial knowledge.
Trust signals
The product shows its work.
In finance, a good-looking answer is not enough. YoungStocks is designed to reveal the evidence, refuse weak requests, and expose the operational layer.
Grounded
Every answer is expected to cite the data behind it.
Replies are tied to concrete market, fundamentals, social, and news sources so users can inspect where the conclusion came from.
Guardrails
When confidence is low, the system refuses clearly.
Instead of bluffing, the assistant explicitly declines questions it cannot answer reliably with the sources currently available.
Privacy-safe operations
Admins can observe usage and lock down access fast.
Usage events, response duration, token consumption, and Google-account blocking are built into the operating surface.
Production thinking
Operational concerns are part of the product design.
Latency
Response duration is logged for performance tuning, cost auditing, and regression detection.
Cost
Model choice is designed to balance quality and price, with token usage tracked for budget oversight.
Security
Google sign-in plus account-level blocking lets admins stop abuse without shutting down the service.
Failure handling
Granular exception capture, timeout boundaries, and traceable logs help the system fail cleanly.
Evaluation roadmap
An asynchronous evaluation pipeline is planned around LLM-as-a-judge, human audit, and user feedback.
Privacy and scale
Per-user memory isolation is in place, with future work on PII scrubbing and dynamic model routing.
See the system live
Open the app and inspect the product directly.
YoungStocks is already online as a working web application. Sign in to explore chat, trade analysis, and the admin-only research surfaces.