About AIvestor

Why was this project created?

I am not a financial advisor nor a fund manager. I am an engineer who, after years of personal market experience, asked a simple question:

Can investing be calmer, smarter, and more consistent – without emotions?

I am not looking for a “magic algorithm” or a guaranteed profit.
I am looking for a way for a system to understand my investment intent, act when I am absent, and make decisions consistent with what I personally consider reasonable.

What AIvestor is (and isn’t)

AIvestor is:

  • a research experiment (R&D), not a commercial product,
  • a decision architecture based on agents and events,
  • a demonstrator of decision intelligence technology, not a promise of profit.

AIvestor is not:

  • financial advice,
  • a black box that “knows better”,
  • a ready-to-use robo-advisor for mass adoption.

What am I testing (hypotheses)

  1. Decisions and context: can technical analysis, ESPI news, market sentiment, decision history, and user strategy be merged into one consistent decision?
  2. Explainability: can every decision be explained and archived in natural language?
  3. Human intent: can an investor express a strategy in natural language, and the system enforce it automatically?
  4. Architectural resilience: is the multi-agent, event-driven approach stable in a real (even if simulated) environment?

Architecture in brief

  • Multi-agent: specialized agents (news, sentiment, instrument, portfolio) instead of a single monolithic model.
  • Event-driven: decisions triggered by events (NEWS_TRIGGER, TAKE_PROFIT_HIT, POSITION_CLOSED).
  • LLM in the loop: large language models (Gemini/GPT) process context and generate recommendations.
  • UI and explainability: web UI shows market data, portfolio, recommendations, and decision history.

What makes AIvestor unique?

  • Non-standard multi-agent architecture – multiple specialized agents collaborating, instead of one black-box model.
  • Human-in-the-loop – the system always operates within the investor’s intent and defined limits.
  • Explainability and audit trail – every decision comes with reasoning and is logged.
  • ~30 data sources – proprietary scrapers + industry portals, financial reports, price data, ESPI news.
  • Conscious experiment, not a product – testing the process, not chasing profit.
  • Challenging test environment – GPW (Warsaw Stock Exchange) with low liquidity as a proving ground.
  • Always-on, event-driven – the system runs continuously, reacting to events in near real-time.

Data and context

AIvestor actively builds its own decision context:

  • pulls raw price data across intervals and performs technical analysis,
  • retrieves company profiles, financial reports, ratios, and valuations,
  • integrates industry news and ESPI announcements,
  • uses approx. 30 scrapers and integrations from various trusted data sources,
  • fetches data on-demand – only when an event requires a decision.

This is the key difference from ChatGPT: here you know what, where, and when was used in making a decision.

Modes and user control

  • Modes: automatic, semi-automatic, manual – configurable via UI.
  • Preselection filters: market, minimum turnover, excluding illiquid “plankton”.
  • Strategy in natural language – global, portfolio-level, and per-instrument.
  • Risk limits: min/max exposure, max SL, etc.
  • Position management: open, reduce, close, extend (add to) if strategy and limits allow.
  • “On-demand” analysis and manual purchases – user can fully take over.

Scope of testing and environment

  • Market: GPW as the starting point – fast access to data, familiar context.
  • Environment: demo account connected to broker API (real-time data, simulated capital).
  • Positions: intentionally small, often illiquid – to generate as many test cases as possible.
  • Style: average position lifetime ~30 days (not day-trading, not buy-and-hold).

The system is tested in tough conditions by design – if it works here, scaling is an engineering, not a conceptual, challenge.

Limitations

  • This is not a product, not advice, and not a profit guarantee.
  • Results on a simulated account do not directly reflect large-scale liquidity or capital.
  • GPW is shallow and volatile – but that’s a deliberate proving ground.
  • Broker integrations may change cousing (temporary) mis-integration.

What’s next?

  • Further testing in R&D mode.
  • Presentations and discussions.
  • Potential partnerships.

AIvestor is a research project.
The content on this site does not constitute investment advice or recommendations.
No results are guaranteed.
All investment decisions remain at your own responsibility.

AIvestor research project – real data, real decisions, real curiosity.