Financial research · AI agents
Build an agent that researches, doubts and verifies before answering.
Aleph turns a complex financial question into a research plan, executes each step against market data and checks its own work.
StatusActive development
My roleArchitecture and lead development
ContextExtended open-source foundation
The terrain
Three real constraints
- 01A useful answer depends on fresh data and traceable sources.
- 02An autonomous agent can loop, drift or conclude too early.
- 03Its history must reveal every tool call and every correction.
The turning point
The decision that changed the project
Treat verification as a working loop, not a final step. Planning, execution, reflection and evaluation became separate, observable and bounded components.
The system
Enough to understand. Not enough to copy.
01Complex question
02Research plan
03Tools and data
04Evaluate and correct
What was built
Real assets, not a concept.
- A 163-file, 38,400-line opportunity-detection engine.
- A market-data collection layer, learning service and on-chain research pipeline.
- Evals, loop limits and JSONL scratchpads that expose the operational reasoning trail.
Observable proof
Transferable lesson
What this experience brings to your project
I do not simply ask an agent to be smarter. I make its mission testable, its context visible and its failures recoverable. That is how an impressive chatbot becomes a working system.
Want to make your agents genuinely reliable?
Discuss my project · free 20 min