A trading journal that prices your mistakes

The challenge

The costly part of trading is not the losing trades — it is the repeated behavioural mistakes a broker's P&L screen has no way to show: the FOMO entry, the revenge trade, the stop quietly moved. A trader can feel the pattern and still never see what it costs, because nothing puts a number on it.

The obvious fix — keep a journal — dies inside a week. Copying every fill from a broker screenshot into a spreadsheet is exactly the manual chore that gets skipped on the first busy day and never resumes. Any tool that depends on that discipline is building on sand.

The approach

Import is a screenshot, not a spreadsheet. A trader uploads a broker screenshot or CSV — Zerodha, Groww, Upstox, Angel One — and an LLM extracts symbol, quantity, and entry and exit prices. Values the model is unsure about are flagged for a human to confirm rather than silently guessed: the same human-in-the-loop, schema-validated pattern we argue for across our AI work, where the machine does the tedious extraction and a person owns the truth.

Mistakes get priced. The trader tags a trade with the rule it broke, and the system computes what that violation cost in rupees — turning “I know I shouldn't revenge-trade” into a line item with a number attached. Over a month, those tags total into the trader's real leak.

Coaching is grounded, not generic. A weekly letter, written by an LLM strictly from that trader's own ledger, names what worked, what leaked, and one process fix for the week ahead — advice that cites the trader's actual trades instead of restating trading platitudes.

The result

Because logging a trade is a thirty-second screenshot upload, the habit survives long enough to compound — which is the whole point of a journal. The trader stops guessing at their weaknesses and starts seeing the rupee cost of each one.

It is also the honest version of the AI pitch we make to every client: extraction handled by a model, judgement kept by a human, and every automated sentence traceable back to the data underneath it.