Case study 03
AI workflows · SaaS
AgentSDR
A multi-tenant cold-email platform that finds and enriches leads through Apollo, ingests website context, generates campaign messages, provisions mailboxes, processes Gmail replies and bounces, synchronizes Smartlead campaigns, classifies reply intent, and updates CRM state.
My role
Primarily architected and built the platform with the engineering team, owning major product areas across onboarding, knowledge workflows, AI outreach, email delivery, mailbox operations, and LLM observability.
Primarily architected and built by Jivesh in collaboration with the engineering team across multiple repository migrations.
Core technologies
System path
A simplified view of the product workflow.
- 01
Lead discovery
- 02
ICP qualification
- 03
Knowledge ingestion
- 04
AI sequencing
- 05
Mailbox events
- 06
CRM state
01
Product overview
Lead discovery and campaign workflow
- 01Apollo-powered lead and company discovery and enrichment.
- 02Ideal-customer-profile configuration and website knowledge ingestion.
- 03AI-generated pitches, email sequences, calls to action, and message editing.
- 04Reply-intent classification, follow-up planning, and CRM state transitions.
- 05Timelines, notes, snoozing, and do-not-contact workflows.
02
My work
Mailbox and delivery infrastructure
I built domain provisioning, mailbox management, warm-up and sending queues, Gmail watch and history processing, inbound reply handling, bounce handling, inbox organization, and Smartlead campaign synchronization.
03
Observability
Tracing AI generation and classification
Langfuse tracing records the prompts, inputs, outputs, and execution paths behind outreach generation and classification. That made debugging, review, and prompt iteration part of the normal engineering workflow.