BACK TO WORK
MithilaStack2024 - PresentLead Full-Stack & AI Engineer

// Neurik

Business automation platform combining AI voice agents with a full CRM-style lead management system to scale B2B/B2C sales and support operations — enabling real-time outbound/inbound calling with transcript capture, call summaries, and human handoff.

The Challenge

Businesses struggle to handle high-volume lead qualification and support calls with human agents, leading to slow response times, inconsistent follow-ups, and missed conversion opportunities.

The Solution

Architected a multi-tenant AI CRM platform with a LiveKit-based Python agent runtime for scalable voice automation, built a modular knowledge-ingestion pipeline processing website and PDF data into structured context, and implemented BullMQ + Redis queue-driven processing for long-running ingestion and call workflows.

// Key Impact Metrics

0s
Call Response Latency
0%
Transcript Accuracy
0+
Concurrent Calls

// Tech Stack

PythonLiveKitAzure OpenAIQdrantBullMQRedisFastifyNode.js

Key Learnings

01.

Low-latency voice interactions require a Python/C++ agent runtime — Node.js event-loop latency is too inconsistent for real-time audio pipelines.

02.

Multi-tenant CRMs need strict data isolation through organizational scoping at the query level, not just at the API layer.

03.

Vector retrieval with source-level traceability significantly improves AI response trustworthiness in enterprise deployments.