BFSIHealthcare· Custom AI agents

SupportIQ — Enterprise AI support agent

<200ms

cache hits

4-route

intent routing

DPDP

compliant

System Architecture

User Query

Presidio PII Scrubber

SPLADE Intent Router

Agentic RAG / Escalation

LLM Synthesis

Presidio Output Scrubber

User

Engineering Deep Dive

Hybrid RAG Approach

Combined SPLADE sparse retrieval with dense embeddings over Qdrant to handle domain-specific financial acronyms that standard embeddings miss.

Sub-process Isolation

Implemented strict LangGraph state machines to ensure that transactional queries could never access conversational endpoints, enforcing security by architecture.

Semantic Caching

Used Redis semantic caching at a 0.85 cosine threshold to instantly serve repeated tier-1 queries, achieving <200ms response times and slashing LLM API costs.

Business Impact

Successfully automated 60% of Tier-1 support tickets while maintaining strict DPDP and GDPR compliance via local PII redaction.

Technologies Used

LangGraph

Qdrant

SPLADE RAG

Presidio

Redis

FastAPI

Groq

Want results like these for your business?

30 minutes. Free. I'll tell you upfront if it's not a fit.