I have a list of user KYC Documents in pdf. After converting pdf to raw .txt file, and when using chatGPT to detect document type I get correct Document Type. However, when using API it incorrectly identifies “consumer cir” as “voter id card”. Multiple runs are producing the same result. Here’s the prompt.
docTypePrompt = “Translate into english. Is the Document Type exactly one of the following: ‘aadhar card’, ‘pan card’, ‘voter id card’,
‘ration card’, ‘driving licence’,‘land tax receipt’,‘property tax receipt’,‘marriage certificate’,
‘death certificate’,‘professional certificate’,‘electricity bill’, ‘telephone bill’,‘gas bill’,
‘bank statement’,‘salary statement’,‘insurance certificate’,‘income tax return acknowledgement’,
‘rc book’,‘consumer cir’,‘msme certificate’,‘shops establishment certificate’,‘police verification certificate’,
‘cancelled cheque’?
Return only the exact Document Type as described above with the exact correct spelling as listed. Get only Document Type and nothing else in JSON format.”
Why this inconsistency for only Consumer CIR report ? All other documents e.g. social security card, electricity bill, driving licence etc. both chatGPT and API return exactly same correct result.