Credit Intelligence
Kalinaw Bank · AI financial analyst

Analyse Marco Reyes

Six sources become a profile, needs, ranked products, a credit scorecard and a note for the credit officer. Code decides every category check, rule and score; models read messy data, write and explain.

MR
Marco ReyesCIF 4410-2207
Salaried professional who revolves a card balance · Makati City
Project manager at Lumen Shared Services Inc. (BPO), regular employee since 2019
87 source rows16 card rows 4 protected fields withheld

Seven steps

  1. 1

    Gather sources

    Code: Loads six sources and withholds protected fields before anything else runs
    AI: No model
  2. 2

    Categorise transactions

    Code: Settles rows with a known transaction code, then checks every model answer against the taxonomy
    AI: Names merchants and picks categories, in parallel batchesgoogle/gemini-2.5-flash-lite
  3. 3

    Build the profile

    Code: Computes income level and stability, essential vs discretionary spend, savings rate, DTI, utilisation and buffer
    AI: Writes the plain-language summaryanthropic/claude-haiku-4.5
  4. 4

    Detect needs

    Code: Runs one rule per need and adds any need the model leaves out
    AI: Picks needs from a fixed list and cites evidenceopenai/gpt-4.1-mini
  5. 5

    Recommend products

    Code: Applies eligibility and suitability rules to the whole catalog
    AI: Ranks only the eligible products and explains the fitopenai/gpt-4.1
  6. 6

    Score credit risk

    Code: Computes the scorecard: points per factor, reason codes, band
    AI: Explains the score; it cannot set or change itanthropic/claude-haiku-4.5
  7. 7

    Officer insight

    Code: Drops claims without a valid source; scans for protected terms and decision language
    AI: Writes the note, then Jev grades itanthropic/claude-sonnet-4.5typesafe/jev-1.13