SKAPIAN LAB◆ CASE STUDY · 13 — ICICI · DIGITAL AUTH
Onboarded in seconds.
An OCR + NER pipeline that reads any ID document across 240+ countries, plus an internal payment-bank app with fraud detection on amount, timing and counterparty patterns.
240+
Countries supported
OCR+NER
ID extraction
NEFT/RTGS
Payment integrations
Docker
Production
ICICI · Digital Auth
● LIVE · ONBOARDING IN PROGRESS
PASSPORT · 🇮🇳 INDIA
NAMEANURAG SHARMA
DOB12 / 03 / 1985
DOCP1234567
P<INDSHARMA<<ANURAG<<<<<<<<<<<<<
✓ EXTRACTED · 6 FIELDS · 200ms
— 01 / OCR + NER EXTRACTION
One document. Six structured fields.
OCR reads the pixels. NER turns raw strings into named fields — works across passport, Aadhaar, driving license, and 240+ international IDs.
— 02 / 240+ COUNTRY SUPPORT
Same pipeline. Any passport.
Document templates and field-mapping rules for every issuing authority — from G20 nations to small island states. MRZ standards, ISO country codes, and locale-aware NER.
IND
USA
GBR
DEU
FRA
JPN
CHN
BRA
RUS
AUS
CAN
ITA
ESP
NLD
CHE
SGP
KOR
ARE
SAU
ZAF
MEX
ARG
CHL
COL
PER
IDN
THA
VNM
PHL
MYS
EGY
NGA
KEN
ETH
MAR
DZA
TUR
ISR
IRN
IRQ
POL
SWE
NOR
DNK
FIN
AUT
BEL
PRT
GRC
HUN
NZL
IRL
CZE
SVK
UKR
BGR
ROU
HRV
SVN
SRB
LKA
BGD
NPL
PAK
AFG
UZB
KAZ
UGA
TZA
GHA
CRI
CUB
DOM
PAN
JAM
HND
GTM
SLV
NIC
VEN
+ 160 more
240 / 240 ISO-3166 codes covered · MRZ + non-MRZ documents · locale-aware
— 03 / LIVE FRAUD MONITOR
Risk scored. Frozen on flag.
Internal payment-bank management with rules + ML — fraud signals tuned on amount, timing, and counterparty patterns. NEFT and RTGS flows watched end to end.
FRAUD MONITOR · LIVENEFT · RTGS · IMPS
1,840 scored / hr● 3 flagged
TIMEAMOUNTCOUNTERPARTYRAILRISKACTION
14:32:08₹ 12,400IDFC · KA · regularNEFT0.12PASS
14:32:14₹ 84,200HDFC · MH · regularNEFT0.18PASS
14:32:17₹ 4,80,000AXIS · DL · newRTGS0.62REVIEW
14:32:23₹ 18,900SBI · TN · regularNEFT0.08PASS
14:32:31₹ 9,90,000KOTAK · WB · midnightRTGS0.91FREEZE
14:32:38₹ 2,400PNB · UP · regularIMPS0.04PASS
14:32:45₹ 2,40,000YES · KA · velocity↑NEFT0.74REVIEW
RULES· amount > 5L on RTGS· velocity > 3× baseline· off-hours & new counterparty· geo mismatch · device ID anomaly
— 04 / ONBOARDING PIPELINE
Document to account. End to end.
Capture
Mobile · branch
OCR
Pixels → text
NER
Strings → fields
Verify
MRZ checksum · live
Fraud Check
Rules + ML
Onboarded
NEFT / RTGS enabled
— 05 / IN PRODUCTION
Onboarding in seconds. Fraud in milliseconds.
240+
Countries supported
6
Fields extracted per ID
<200ms
Doc → fields latency
NEFT/RTGS
Rails watched
"Open-source NER, packaged for production. Fraud detection tuned on amount, timing, and counterparty patterns — the kind of work that has to survive audit."
— On what production-grade banking AI looks like
— 06 / BUILT WITH
The toolkit.
HuggingFace OSS
Docker
Financial APIs
SQL
Weblogic
— NEXT CASE STUDY