Our work / AI product

BAQI AI: a bank statement in, a halal portfolio out

Most people in Pakistan do not invest because they lack clarity, guidance and trust. BAQI AI reads a bank statement, finds the investable surplus and builds a personalised, Shariah compliant PSX portfolio in under a minute.

85%directional accuracy on PSX forecasts
ClientFintech, Pakistan
What we didLead AI systems engineering
StackCrewAI, Claude, LangChain, Supabase, Machine learning

THE PROBLEM.

People did not know how much they could afford to invest, had no simple way to analyse PSX stocks and could not trust that recommendations were Shariah compliant. Existing tools were too complex, too generic or ignored Islamic finance.

WHAT WE BUILT.

  • Claude parses bank statements and extracts 14 behavioural spending signals
  • A BAQI engine that classifies spending as fixed, discretionary or leaks and calculates the investable surplus
  • Five CrewAI agents on Claude: spending analyst, risk specialist, market analyser, Shariah officer and strategist
  • Dual layer Shariah screening against KMI 30 and Meezan Bank
  • A 21 day PSX forecasting ensemble of SVM, MLP, gradient boosting and ridge models
  • Wavelet denoising that cut forecast error by 30 to 42 percent

WHAT CHANGED.

  • A personalised, Shariah compliant portfolio in under 60 seconds
  • 85% directional forecast accuracy on Pakistani stocks
  • Recommendations with tickers, PKR amounts and a clear rationale
Next case studyWorkout plans that match what you can actually lift