Skip to content
woodfrog

·

AI
Trading infrastructure, NSE, BSE and MCXCustomer story, named

Faster research and documentation.

On a platform built for ultra-fast trading, every strategy still waited for a person to write up the backtest. Now every run writes its own summary, risk commentary and explanation, and research and documentation time is down by about 60 to 70%.

One backtest, written up automaticallySchematic
Written from structured results only. Raw market data never leaves the engine.
60 to 70%Less research and documentation time (about, a range)
750+Users on the platform, and growing
Same headcountOutput grew without more people
4 monthsA woodfrog pod of three, May to September 2026
WhoStockJarvis, India's trading infrastructure platform for NSE, BSE and MCX
EngagementFour months, May to September 2026, a woodfrog pod of three. Built on Claude, through the Anthropic Claude Partner Network.
BuiltAn AI layer inside the platform's Strategy Lab that writes the analysis of every backtest
Before and after

Every backtest now explains itself.

Before
  • An analyst read complex results, Greeks and payoff profiles by hand
  • Each strategy waited for its write-up
  • Quality varied from analyst to analyst
  • Research output could not scale
After
  • Every run writes a performance summary, risk commentary and explanation
  • Analysts move from raw results to polished insight far faster
  • The same structure every time, so strategies are easy to compare
  • Research scales without a matching rise in headcount
The problem

Research that waited on a person to write it up.

StockJarvis gives users tick-by-tick history, second-by-second simulation and multi-leg basket backtesting. Then the result waited for a person. An analyst had to review the output, interpret the Greeks and payoff profiles, write a clear summary, and turn it into insight for users.

What was broken
  • Complex results were read and interpreted by hand.
  • Each strategy waited for its write-up, so decisions took longer.
  • Documentation quality varied, which made it harder to trust and compare.
  • Research and documentation stayed heavily manual and could not scale.
You might be here if
  • Your analysts read complex backtest results by hand.
  • Each of your strategies waits for its write-up.
  • Your write-ups vary from analyst to analyst.
  • Your research output cannot scale without more people.

Why it mattered. In a market where accuracy and quick checking of what-if scenarios decide things, these delays held back growth and user confidence.

The approach

Make the write-up an output of the backtest.

We stopped treating documentation as a job that follows the backtest. After every run, the system now produces a structured analysis. We also drew a boundary: Claude works only from the engine's structured results, never from raw market data. It runs alongside the trading engine, not in the path of any trade.

The platform
Strategy LabBacktests and simulations
Structured resultsGreeks, payoffs, win rates, drawdowns
woodfrog built
Secure connectorsStructured results only
Research agents on ClaudeRead multi-leg strategies
Two layers of safeguardsPrivacy first, then accuracy
What each run now produces
Performance summary
Risk commentary
Adjustment logic explained
60 to 70% less time
Deliberately not doneRaw market data never reaches Claude, and nothing here places a trade
Claude works from structured results only.schematic
The solution

Connectors, agents and safeguards inside the platform.

What was built
  • Secure, authenticated connectors that pull structured backtest results without exposing raw market data.
  • Prompt frameworks so Claude reads complex trading logic, entry and exit rules and risk limits correctly.
  • Research agents that understand multi-leg strategies and write precise technical summaries for every backtest run.
  • Two layers of safeguards: privacy, access control and audit logging first, then accuracy checks and professional, controlled wording.
  • Questions in plain language about past performance, drawdowns and win rates, plus insight cards refreshed with the latest checked numbers.
  • Live in production with monitoring, alerting and fallback, tested across NSE, BSE and MCX strategy types.
The final month ran 10 August to 10 September 2026 in three steps: the secure data layer first, the research agents second, hardening last.
How it ran, four months
The impact

Faster research, with the same team.

Research and documentation time, before and afterBefore is set to 100. After is shown as a band, because the result is a range.
60 to 70%less time
Hatched part is the rangeIndexed to 100. No before or after durations are published.
Proof it is in use
BacktestsEvery live run produces its own write-up
StrategiesDocumentation generated for live strategies
Daily useBy the research and product teams
FeedbackPositive, as summarised by the client
Get your research writing itself.A short call, or record a message. A person replies within one business day.
Contact usCertified Services Partner, Claude Partner Network