Vision Tasks

AI agents that analyze images and documents from customer conversations.

Active vision agents
4
1 in draft
Images analyzed today
131
Images analyzed (7d)
650
Avg accuracy
92.3%
Avg latency
2.4s
per analysis
Cost (7d)
₹12,840

Card Damage Inspector

v1.4
activeGPT-4o

Analyzes uploaded card photos, classifies damage type (bent, cracked, magnetic strip, chip), and recommends action (replacement / re-issue / no action needed).

last 3 analyzed
PII auto-masked
28
today
142
7d
91%
accuracy
Top detections
Bent corner
38%
Magnetic strip wear
24%
Chip damage
18%
Used in flows
Card Replacement RequestBlock Lost Card
by Aman Guptalast used 2m ago

Cheque Vision

v1.0
activeClaude 3.7 Sonnet

Reads cheque images and extracts MICR code, payee name, amount in figures and words, date, signature presence. Flags discrepancies between figures and words.

last 3 analyzed
PII auto-masked
14
today
67
7d
88%
accuracy
Top detections
Amount mismatch
12%
Date issues
8%
Missing signature
4%
Used in flows
Cheque Status CheckMobile Cheque Deposit
by Priya Menonlast used 11m ago

KYC Document Reader

v2.1
activeSarvam Vision

OCRs and verifies KYC documents (Aadhaar, PAN, Passport, Driving License, Voter ID, Utility Bills). Cross-references with provided customer data and flags mismatches.

last 3 analyzed
PII auto-masked
67
today
287
7d
94%
accuracy
Top detections
Aadhaar
52%
PAN
28%
Passport
12%
Used in flows
KYC Re-verification SubmissionAccount Opening Onboarding
by Sneha Patellast used just now

Selfie + Liveness Verification

v1.2
activeGPT-4o + Hyperverge

Captures customer selfie via WhatsApp/Web Chat, runs liveness check, matches against Aadhaar/PAN photo for video KYC. Integrated with Hyperverge.

last 3 analyzed
PII auto-masked
22
today
142
7d
96%
accuracy
Top detections
Match success
89%
Liveness fail
7%
Low light retake
4%
Used in flows
Video KYC
by Rahul Iyerlast used 5m ago

Insurance Claim Damage Assessor

v0.3
draftGemini 2.0 Vision

Vehicle damage assessment from claim photos, estimates severity and category. Drafted for the Acko workspace.

last 3 analyzed
PII auto-masked
0
today
12
7d
78%
accuracy
Top detections
Bumper damage
41%
Side panel dent
26%
Glass crack
18%
Used in flows
Motor Claim Intake
by Sneha Patellast used 2d ago

Sample image gallery

PII auto-masked per privacy policy

Recent damaged cards detected

8

Recent KYC documents processed

8

Recent cheques scanned

6

Edge cases for review

7

Flows using vision steps

Tap to open the flow editor
Card Replacement Request
uses Card Damage Inspector
Cheque Status Check
uses Cheque Vision
KYC Re-verification Submission
uses KYC Document Reader
Account Opening Onboarding
uses KYC Document Reader + Selfie Verification
Mobile Cheque Deposit
uses Cheque Vision
Motor Claim Intake
uses Insurance Claim Damage Assessor

Accuracy & quality

Last 30 days
Card Damage Inspector
91%-1.0
Cheque Vision
88%-1.0
KYC Document Reader
94%-1.0
Selfie + Liveness Verification
96%-1.0
Top miss categories
  • Chip-card classification14 misses
  • Cheque amount in words9 misses
  • Aadhaar address line 27 misses
  • Selfie low light6 misses
Drift alert

Card Damage Inspector accuracy dropped 3.4% in the last 7 days, primarily on chip-card classifications. Recommended: add 50 chip-card samples to the training set.

Inter-rater agreement (labeled samples)
87% κ = 0.74

Compliance & retention

Mask Aadhaar number in stored copies
Mask PAN number in stored copies
Blur signatures in stored copies
Blur faces in stored copies
Customer consent before processing
DPDP-aligned erasure on request
Retention
Non-KYC images30 days
KYC images (RBI)7 years
StorageAWS Mumbai (encrypted)