standpulse customerpro crystalai writing
Case Study · Fintech · AI Automation

Lending Automation

End-to-end AI pipeline from lead scoring to disbursement decision.

orchestrated flow · 6 modules
1

Propensity Algorithm

ml · ranks applicants by conversion

0.82 high intent

conversion-likelihood rank

6

Orchestrator Agent

n8n · coordinates modules

in · propensity out · Camunda

orchestrates modules 2 · 3 · 4 · 5

2

Voice AI

outbound calling agents

connected
objection handled ✓
ElevenLabs Gnani
3

Call Analysis

supabase · llm signals

Caller Team Lead Sales Head
loan stagecredit processing
intenthigh
effectivenessobjection handled
urgencymedium
risk flagsnone
4

WhatsApp AI Bot

document submit · no portal login

Share your PAN & bank statement
PAN.pdf statement.pdf

→ feeds IDP

5

IDP

nestjs · document processor

1checklist · PAN · Aadhaar · bank · GST
2extract · PAN ABCDE1234F 0.98
3checks · 6-mo ✓ names ✓ GST≠PAN ⚠
4CIBIL · consent + OTP
5hand-off → Camunda
validated case → Camunda decision engine
6 autonomous modules · 1 orchestrator OpenAI Anthropic (managed) Together AI (OSS + embeddings) self-hosted pgvector (RAG) NestJS Supabase n8n ElevenLabs / Gnani Camunda
Case Study · Security · AI Automation

VAPT Automation

Full vulnerability assessment pipeline. Recon to report. Entirely automated.

target acme.example · authorized
domain acme.example · go-ahead ✓ granted
↻

Planner Loop

crewai · goal-directed

1choose next task
2run tool
3read result
4judge · re-plan ↻

queries KB · MITRE · runs scans in parallel

1

Recon & Scan

endpoint discovery

nmap
ferox
zap
wappa
stack detected · PHP · Apache
2

Detection

mvp focus

SQL Injection · candidate
OS Command Injection · candidate

detected · awaiting exploit approval

human approval
3

Exploitation

gated · human-approved

▶SQLMap → SQLi
▶Commix → OS cmd inj
Exploit-DB · keyword → filter by stack py2 → modernize → py3 · inject args
interactive code writer $ run exploit.py stderr: SyntaxError (py2) ↳ auto-fix · retry $ running ✓
validated on OWASP Juice Shop + DVWA · illustrative run
CrewAI AWS Bedrock (profile-based LLMs) MITRE knowledge base embedding pipeline over scan results manual baseline ~3 months objective (goal) 1–2 days fastest agentic run 23 min · single goal (exploit → hostname) parallel run 15 min · single goal (DVWA hostname) validated on OWASP Juice Shop + DVWA code-gen evaluated · DeepSeek Coder (best) · Claude 3.5 v2 · Yi Coder · Llama Code