How VerbaFlo Revolutionizes Real Estate Customer Interactions with Conversational AI Agents and Zilliz Cloud
Up to 50%
Cost Reduction vs. MongoDB infrastructure
Sub-10 millisecond
Performance Breakthrough
60+ QPS
production loads with room for growth
Superior multi-tenant data
isolation and partitioning
It has saved us cost by at least 25 to 30% on what we were earlier incurring on the vector database, with potential savings reaching up to 40-50% during peak traffic periods.
Rachit Jindal
About VerbaFlo
VerbaFlo is among the world's most adaptable conversational AI platforms, revolutionizing how real estate professionals manage operations and interact with clients. Their AI-driven platform unifies voice, chat, text, and email communications into a single intelligent engine that delivers near-human conversations 24/7. With advanced features like AI-powered client matching, dynamic pricing engines, and automated property management, VerbaFlo drives consistency, visibility, and measurable performance across every customer interaction. With plans to expand into the automotive industry, VerbaFlo is positioned as the industry leader in AI-powered communication solutions.
The Challenge: Scaling Beyond Infrastructure Limitations
As VerbaFlo's AI platform gained traction with real estate customers, their initial MongoDB Atlas infrastructure began showing critical limitations that threatened both customer experience and business growth. The company was operating at production query loads exceeding 50-60 QPS, pushing their system beyond what MongoDB could reliably handle.
Query latencies were too slow for real-time conversational AI, where users expect instant responses, especially during voice calls, where delays become immediately noticeable. The team had experimented with various vector databases, including Qdrant and Pinecone, but none could scale to meet their demanding performance requirements.
"We were quite stuck at a point where our production environment was having a QPS over 50 or 60, and most of the existing databases were not able to scale to that level. Even MongoDB Atlas M30 or M40 was not able to scale to that point," explains Rachit Jindal, Senior AI Engineer at VerbaFlo.
The challenges extended beyond performance. VerbaFlo faced four primary concerns: cost inefficiency with MongoDB's M40 clusters being too expensive for their requirements, inadequate high concurrency support, poor query times, and limited partitioning capabilities. The team required sophisticated data isolation for their multi-tenant architecture; however, MongoDB was unable to provide proper namespace isolation for different collections and environments.
These infrastructure limitations were directly impacting VerbaFlo's ability to deliver seamless, real-time property management automation. With aggressive growth projections and plans to scale their entity count by 10-20x within a few quarters, the company needed a solution that could grow with their ambitions.
The Journey to Zilliz Cloud
Recognizing that their database architecture had become a critical business constraint, VerbaFlo's engineering team conducted a comprehensive evaluation of vector database alternatives, including Zilliz Cloud (a managed service of the open-source Milvus), Pinecone, Qdrant, PostgreSQL with vector extensions, and self-hosted Milvus on AWS.
Zilliz Cloud emerged as the clear winner for several compelling reasons. Rachit had previous experience hosting Milvus on-premises, which gave him confidence in its horizontal and vertical scaling capabilities, as well as its internal architecture. "Based upon the previous experience I have with Milvus, its internal architecture around scalability, horizontal and vertical scaling, it's definitely the go-ahead for our team to consider it as one of the potential databases," he notes.
The evaluation revealed that Zilliz Cloud offered sophisticated partitioning and isolation capabilities that perfectly matched VerbaFlo's multi-tenant requirements. Unlike MongoDB, where they had "very limited capability for creating an index in a particular collection" and couldn't "define a proper namespace in a proper database when creating a collection," Zilliz Cloud provided proper database and namespace isolation for clean separation between production environments, staging systems, and their customer-facing AI playground.
Another decisive factor was Zilliz Cloud's built-in re-ranking algorithms. Instead of developing custom algorithms, VerbaFlo could leverage weighted re-ranking and RRF (Reciprocal Rank Fusion) systems out of the box. "With Zilliz Cloud, there were certain internal algorithms that are quite suited to the use case that we require, like the weighted re-ranking and RRF re-ranking systems. That helped us to directly use their SDK as well as compute them on the fly so that it can reduce our query time," Rachit explains.
Zilliz Cloud’s native support for hybrid search—combining semantic and keyword-based queries—perfectly aligns with VerbaFlo's query expansion pipeline. Most importantly, Zilliz Cloud could deliver the sub-10 millisecond latency that their real-time applications demanded, while also offering better managed services compared to other alternatives.
How Zilliz Cloud Powers Real-Time Conversational AI Agents at Scale
VerbaFlo's previous MongoDB-based architecture struggled to handle production loads exceeding 50-60 QPS, resulting in bottlenecks that prevented real-time responses. With Zilliz Cloud as the foundation, they've built a sophisticated conversational AI platform that currently handles over 40,000-50,000 entities with seamless performance.
The New Agent Architecture with Zilliz Cloud
At the core of VerbaFlo's system is an intelligent routing mechanism that classifies user intents and directs queries to specialized processing nodes. Zilliz Cloud powers the critical RAG (Retrieval-Augmented Generation) node, which retrieves relevant context from property data, customer histories, and market information to generate accurate, personalized responses.
One of Zilliz Cloud's most valuable capabilities for VerbaFlo is its advanced partitioning system, which enables two critical types of data isolation:
Environment separation: Clean isolation between their AI playground/labs environment and production collections, allowing customers to test and experiment without affecting live operations
Performance optimization: Data segmented into smaller partitions based on specific scenarios, enabling targeted queries of limited sections rather than entire collections
This partitioning approach directly addresses the namespace limitations they faced with MongoDB, where they had "very limited capability for creating an index in a particular collection" and couldn't "define a proper namespace in a proper database when creating a collection."
Powering the Hybrid Search Pipeline with Zilliz Cloud
VerbaFlo's system doesn't rely on simple semantic searches. Instead, their query expansion pipeline creates both semantic and keyword queries for every user request. Zilliz Cloud's native hybrid search capabilities and built-in re-ranking algorithms (weighted re-ranking and RRF) eliminate the need for custom development while delivering superior results.
"With Zilliz Cloud, there were certain internal algorithms that are quite suited to the use case that we require, like the weighted re-ranking and RRF re-ranking systems. That helped us to directly use their SDK as well as compute them on the fly so that it can reduce our query time," Rachit notes.
Seamless Migration: From MongoDB Constraints to Zilliz Cloud Performance
The migration from MongoDB Atlas to Zilliz Cloud demonstrated how a well-planned database transition can resolve critical performance bottlenecks without disrupting business operations.
VerbaFlo's engineering team approached the migration systematically, leveraging Zilliz Cloud's comprehensive documentation and SDK resources. The migration involved transferring their existing vector embeddings and metadata while implementing new partitioning strategies that MongoDB couldn't support.
"The documentation on Zilliz Cloud as well as Milvus was quite detailed and comprehensive. Given that kind of documentation and all the SDK docs, it was pretty much easy to migrate," Rachit recalls.
Results and Benefits
The impact of migrating to Zilliz Cloud was immediate and substantial across multiple dimensions.
Up to 50% Cost Reduction: VerbaFlo achieved immediate cost savings of 25-30% compared to their MongoDB Atlas infrastructure. During peak traffic periods, the potential savings could reach 40-50% thanks to Zilliz Cloud's efficient auto-scaling capabilities and ideal cluster configurations for production-ready environments.
Sub-10 millisecond Performance Breakthrough: Query latencies dropped from 15-20 milliseconds to consistently under 10 milliseconds—a dramatic improvement that customers noticed immediately. This sub-10 millisecond performance enables truly conversational AI experiences across voice calls, chat widgets, and messaging platforms.
Enhanced Scalability: With the ability to handle their current 50-60 QPS load with significant headroom for growth, VerbaFlo now has the infrastructure foundation to support their aggressive scaling plans without performance degradation.
Superior Data Isolation and Partitioning: Zilliz Cloud's advanced partitioning capabilities solved VerbaFlo's complex multi-tenant requirements that MongoDB couldn't handle, significantly improving both security and performance for their multi-tenant architecture.
Compliance and Security: Zilliz Cloud's GDPR compliance and security certifications enable VerbaFlo to confidently serve customers in the UK and EU regions, supporting their international expansion plans.
Improved System Reliability: With 99.5%+ uptime, Zilliz Cloud supports VerbaFlo's 24/7 service commitments, critical for real estate applications where customer inquiries don't follow business hours.
Seamless Ecosystem Integration: Zilliz Cloud's native compatibility with cloud infrastructure providers enables VerbaFlo to leverage AWS services, including ECS, EKS, and S3 buckets for hosting and deployments, while supporting their multi-cloud strategy that spans Google Cloud and various AI service providers.
Future Plans: Expanding AI Innovation Across Industries
With their infrastructure challenges resolved, VerbaFlo is pursuing ambitious expansion plans across multiple dimensions. The company projects significant growth in entity count, expecting to scale 10-20x within the next few quarters as traffic increases and they onboard more customers to their AI playground feature.
"The team is in a constant phase of developing, optimizing our existing flows, and implementing new flows internally into the system," Rachit explains. "We definitely desire to actually leverage Zilliz Cloud to suit some of the use cases we're planning in the future."
More significantly, the solid foundation provided by Zilliz Cloud has enabled exploration of entirely new industry verticals. VerbaFlo is actively developing AI agent architectures for the automotive and pharmaceutical industries, leveraging the same conversational AI capabilities that have proven successful in the real estate sector.
The technical capabilities that make this expansion possible include Zilliz Cloud's international compliance standards and reliable performance at scale, providing VerbaFlo with the confidence to think beyond its current market and build for global growth across multiple sectors.
Conclusion
VerbaFlo's partnership with Zilliz Cloud demonstrates how the right infrastructure choice can transform technical constraints into competitive advantages. By achieving 25-30% cost savings, sub-10 millisecond query performance, and robust scalability with Zilliz Cloud, VerbaFlo has positioned itself for aggressive growth across multiple industries.
As VerbaFlo continues to innovate in f and expand into new markets, Zilliz Cloud provides the scalable foundation they need to transform how businesses interact with their customers.
"The sub-10 millisecond latency for queries is quite the benchmark in the industry for vector databases," Rachit concludes. "Combined with the cost savings and reliability, Zilliz Cloud has become a strategic enabler for our broader vision of transforming conversational AI across industries."
- About VerbaFlo
- The Challenge: Scaling Beyond Infrastructure Limitations
- The Journey to Zilliz Cloud
- How Zilliz Cloud Powers Real-Time Conversational AI Agents at Scale
- Seamless Migration: From MongoDB Constraints to Zilliz Cloud Performance
- Results and Benefits
- Future Plans: Expanding AI Innovation Across Industries
- Conclusion
Content
The sub-10 millisecond latency for queries is quite the benchmark in the industry for vector databases. Combined with the cost savings and reliability, Zilliz Cloud has become a strategic enabler for our broader vision of transforming conversational AI across industries.
Rachit Jindal