There is a gap that stops more people than any data limitation or pricing tier: the distance between having an idea and having a strategy you can test. saral.money’s AI assistant exists to close that gap — without taking the strategy out of your hands. This is what it does, and, just as importantly, what it does not.
The blank-canvas problem
“Buy quality companies when momentum turns up.” That is a perfectly good investment thesis. It is also not yet a strategy.
To test it, you have to make a dozen concrete decisions. What is the universe: Nifty 500, or midcaps too? What defines “quality”: return on equity above some threshold, low debt, both? What defines “momentum turning up” — price above its 200-day average, or a positive 12-month return? How many stocks do you hold, and how do you size them? How often do you rebalance?
Each decision is small. Together they are the wall most people never get over. They understand the market perfectly well; they just stall at the blank canvas. The skill of translating a thesis into a precise, testable pipeline is separate from the skill of having the thesis — and it is the one that has gated systematic investing.
What the assistant actually does
You describe the thesis in plain language. The assistant returns a strategy as an editable pipeline — the same visual flow the manual builder uses: a universe node, a fundamental filter, a ranking rule, and a portfolio output, with your rebalance frequency and position-sizing choice set as part of that pipeline.
The word that matters is editable. The assistant does not hand you a sealed result. It hands you a starting structure you can open node by node: change a threshold, swap one factor for another, tighten the universe, delete a rule you disagree with. Take “buy quality companies when momentum turns up”: the assistant might scope the universe to Nifty 500, write a fundamental filter such as ROE > 15 AND PE < 30, and rank the survivors by 12-month momentum. Open that filter node on a name like HDFCBANK and you can see exactly why it passed, or didn’t. It does the scaffolding; you keep the authorship.
That is a deliberate design choice. The alternative — an AI that quietly decides everything and gives you a single “buy this” output — would be easier to build and far less useful. You would have traded one black box (your own intuition) for another (the model’s), and learned nothing.
Why it stays white-box
Everything the assistant produces is a set of explicit, readable rules. There is no opaque model sitting between you and your money. You can read exactly why a stock is in the portfolio: it passed this filter, ranked in this position on this factor, got this weight.
This is not only a usability decision; it lines up with regulation. SEBI’s February 2025 retail algorithmic trading framework treats “white-box” algorithms — those whose logic is disclosed to and controlled by the user — as the lighter-compliance path, distinct from opaque “black-box” systems. A strategy made of inspectable rule nodes is white-box by construction. The AI lowers the barrier to authoring those rules; it does not hide them. More on what the framework actually requires.
What it deliberately does not do
Here is the honest part, because an assistant oversold is an assistant that loses you money.
- It is not a market oracle. It does not predict where prices are going. It translates a thesis into a structure; the thesis is still yours, and a bad thesis well-structured is still a bad thesis.
- It can be wrong. An AI can misread what you meant, or propose a rule that reads sensibly and backtests badly. Generated strategies are starting points, not finished products.
- It does not remove the validation step. Nothing the assistant produces is exempt from the one rule that governs everything on this platform: it has to survive a backtest across 15+ years of data, including 2018 and the March 2020 crash, before it deserves real capital. The AI gets you to a testable strategy faster. It does not get you out of testing it.
Treat it the way you would treat a sharp colleague who sketches a first draft: useful, fast, occasionally wrong, and never the final word.
What to try next
Open the strategy builder and describe a thesis in your own words — then do the important part: open the nodes it generates and read them. Change one factor and see how the backtest responds. The assistant’s job is to get you to that first testable draft in minutes instead of hours. Judging the draft is still yours.