Project 04
Stock Suggestion Agent
The Problem
Generic recommendations are useful to no one in particular.
Most stock screeners give generic output based on technical indicators or broad themes. They don't know your investment thesis, your risk appetite, or what you already hold. A value investor and a momentum trader get the same recommendations — which means the output is useful to neither.
Retail investors who have a clear thesis — "mid-cap Indian manufacturing, low debt, high promoter holding" — have no easy way to translate that into a screened shortlist without knowing how to write screener formulas or navigate the filter UIs that professional tools offer.
The gap isn't knowledge — most serious retail investors understand what they're looking for. The gap is the translation layer between a natural language thesis and a set of quantitative filters that actually runs against real data.
The result: people either fall back on stock tips from social media, or spend hours manually filtering through lists that weren't built with their criteria in mind.
The Solution
Your thesis in. A ranked shortlist out.
An agent where you describe your investment thesis in plain English. The agent parses your intent, maps it to quantitative filters, runs a screen against market data, and returns a ranked shortlist with per-stock rationale tied back to your stated criteria. If your thesis has gaps or internal contradictions, the agent surfaces them before running the screen — so you refine your thinking, not just your filters.
The key design decision: the thesis is the input, not a set of dropdowns. This keeps the tool flexible across investment styles and forces the agent to do interpretation work rather than the user. A value investor, a GARP investor, and a dividend hunter all use the same interface — the agent adapts to them.
Each result in the shortlist includes a one-paragraph rationale explaining exactly why the stock matches the thesis — not a generic summary, but a response to the specific language the user used. This makes the output actionable rather than another list to manually validate.
Product Screenshots