5+ years architecting real-time voice AI agents, LLM document intelligence pipelines, vector DB search systems, and vLLM GPU inference microservices for enterprise workflows.
{
"lead": "Enterprise Client",
"intent": "RFP_Qualified",
"confidence": 0.98,
"action": "SYNC_SALESFORCE"
}Detailed breakdown of architectural decisions, constraints, trade-offs, and measurable outcomes.
Sub-second bidirectional voice pipeline connecting Gemini Live, Plivo telephony, and MiClient CRM.
Architected a real-time voice AI qualification system that ingests inbound phone leads via Plivo, streams bidirectional audio into Gemini Live, and streams structured CRM updates.
Automated PDF & scanned document requirement extraction, product-customer matching, deal prediction, and confidence scoring.
Built an intelligent Request for Proposal (RFP) engine that parses complex multi-page PDF/scanned documents, extracts granular line-item requirements, and predicts deal win probability.
High-throughput document ingestion engine processing 1,000+ documents/month with 15+ third-party CRM integrations.
Architected a scalable document intelligence pipeline processing 1,000+ invoices, purchase orders, and customer contracts monthly while integrating seamlessly with Gmail, Salesforce, QuickBooks, and WhatsApp.
High-performance vLLM GPU cluster, vector embedding pipelines, AWS-to-GCP modernization, and hybrid FastAPI + Node.js services.
Designed and modernized the AI core backend, transitioning workloads from AWS to GCP, deploying self-hosted vLLM GPU inference instances, and building scalable RAG vector search pipelines.
End-to-end signal pipeline engineering: from edge client connections to GPU inference clusters and CRM state mutation.
AI components are probabilistic; backend systems must be deterministic. Pydantic schemas, fallbacks, and retry queues guarantee system integrity.
Tracing every token, audio chunk, and API callback with per-request correlation IDs to pinpoint latency anomalies and model degradation.
Self-hosting vLLM GPU inference instances and deploying hybrid semantic cache layers to keep p95 latency low and token costs predictable.
Automated regression testing on extraction pipelines using synthetic dataset benchmarks before promoting prompt changes or model weights.
Production-tested webhooks, REST APIs, OAuth workflows, and streaming WebSocket pipelines built for MiClient CRM/CPQ platform.
5+ years leading backend & AI architecture across high-growth startups and tech enterprises.
Competitive programming focused on algorithmic optimization, dynamic programming, and graph algorithms.
Problem solving under strict time and space complexity constraints.
Built scalable tech solution addressing real-world problem statement under 36-hour hackathon environment.
Test raw text chunking and structured JSON extraction. Demonstrates client payload parsing with rate-limiting validation.
{ "intent": "RFP_MODULE_ENQUIRY", "integrations": ["QuickBooks"], "score": 0.96 }Whether you're looking for a Tech Lead to architect real-time voice AI, build document intelligence systems, or scale vLLM GPU inference infrastructure, feel free to reach out.