Hritul Srivastava

Distributed Systems & GenAI Engineer for SaaS Startups

For SaaS teams past “it works on staging”: Kafka pipelines that hold at ~20k events an hour, and RAG systems that keep context past turn five.

Featured Work

SAGE-AIlive

A context-aware agentic RAG system that retrieves, reasons, and acts across multi-turn conversations without losing context, backed by MongoDB Atlas Vector Search.

Answers stay grounded across the whole conversation.

RAGAgentsVector Search
Alphonsolive

A high-throughput social listening pipeline ingesting ~20,000 posts/hour from Reddit and social platforms for real-time lead generation.

Lead discovery moved from hourly batches to near real-time.

KafkaHigh-ThroughputLead Gen
Sales Call → CRM Automationwip

A productized pipeline that turns a sales call recording into structured CRM fields — transcript, LLM extraction, and a webhook straight into your CRM.

CRM fields populate before the call ends.

AutomationLLM ExtractionWebhooks

What I Do

01

Distributed Systems & Backend

  • Kafka pipelinesingestion that doesn't fall over at 20k events/hr
  • Redis caching layershot paths that stay fast when traffic spikes
  • High-throughput APIsendpoints that hold under concurrent load
  • Sales call → CRM pipelinesrecordings landing as structured CRM fields
  • Webhook integrationsdeliveries that retry instead of vanishing
  • Observability (OpenTelemetry)traces that show where the latency went
02

GenAI & AI Infrastructure

  • RAG pipelinescontext that survives a multi-turn conversation
  • LLM agentstool calls that fail loudly, never silently
  • Vector search (MongoDB Atlas, Pinecone)relevance that holds as the corpus grows
  • Transcript processingraw call audio to clean, queryable text
  • Structured data extractionoutput that matches your schema every time
  • Evals & monitoringregressions caught before your users find them
Free: RAG eval checklist + distributed systems design-doc template.The two documents I actually use on client projects.