Redefining Career Discovery: How Jobright Uses Zilliz Cloud to Deliver Faster, Smarter Talent Matches Beyond LinkedIn
<50ms
Latency
Zero Downtime
with faster feature deployments
Zero Hassle
for Database Administration
Cost Cut
Per User
Choosing Zilliz Cloud was one of our best early decisions. It enabled us to build the product we envisioned rather than the product our infrastructure limitations would allow. In AI applications, that difference often determines success or failure.
Ethan Zheng
About Jobright: Redefining Job Search in the AI Era
Jobright is an AI-native job search platform that’s rapidly gained ground on industry incumbents like LinkedIn and Indeed. In just over two years, it has reached nearly 100,000 daily active users and now leads the industry in average user session duration, according to SimilarWeb.
Built by engineers Eric Cheng and Ethan Zheng, who left Big Tech roles to fix what they saw as a broken discovery problem. Jobright delivers smarter, more personalized job matching through semantic search, graph analysis, and real-time system feedback. The platform’s standout features, like H-1B sponsorship filtering and referral discovery, have made it especially valuable to international professionals and skilled technical talent.
Why Traditional Search Broke at 2 Million Queries a Day
As Jobright gained traction, its technical demands escalated. The platform’s core value, real-time, personalized job matching—quickly exposed the limitations of traditional databases.
Constantly Changing Data: Job postings change constantly. Over 400,000 postings are added or removed daily. Most databases can’t ingest and delete data at that rate without performance issues.
Semantic, Multi-Dimensional Matching: Jobright doesn’t just match keywords. It searches across job descriptions, skills, career trajectories, and company culture. Every user interaction involves multiple vector searches plus filtering for location, visa status, experience level, and more.
Real-Time Response at Scale: Users average over 40 interactions daily. That translates to over 2 million queries per day, all expected to return sub-100ms results.
Unpredictable Load Patterns: Hiring sprees or layoffs from major employers cause data ingestion and query surges. The infrastructure had to absorb these spikes without compromising performance.
"This isn't like a typical recommendation system where you can batch process overnight," explains Ethan, Co-founder & CTO of Jobright. "Users expect instant results that reflect the latest job postings and their evolving preferences. The infrastructure has to handle both massive scale and real-time responsiveness simultaneously."
How Jobright Benchmarked the Vector Database Market
As Jobright's simple, Python-based matching system began to fail under load, the team faced a critical infrastructure decision. Rather than rushing into a quick fix, Ethan spent an entire week systematically evaluating every major vector database option in the market, including FAISS, Pinecone, Elasticsearch, open-source Milvus, and Zilliz Cloud. His approach was methodical and hands-on: "We don't just read documentation or listen to vendor pitches. We build actual MVPs with each solution and test them rigorously in our specific environment."
Each option revealed critical limitations:
Facebook FAISS initially seemed promising, given its battle-tested performance at Meta's scale. However, when Jobright attempted to handle concurrent queries while constantly inserting and deleting vectors, this Python implementation became unstable, with memory spikes and crashes during peak usage.
Pinecone offered a managed solution, but regional limitations created unacceptable latency for their global AWS-based infrastructure. "We're a global platform serving users worldwide. Having our vector database in only certain regions would have crippled our international user experience," said Ethan.
Self-hosted Milvus performed well with concurrent workloads and demonstrated scalability, but it required significant operational overhead for self-management—overhead that would have diverted critical engineering resources from product development.
Elasticsearch, while reliable for basic operations, cannot efficiently combine vector similarity with Jobright's dozens of filter dimensions, such as visa sponsorship, experience level, and company size.
Then Ethan tested Zilliz Cloud. The difference became apparent within hours—while other solutions required constant tuning and monitoring, Zilliz Cloud handled Jobright's demanding workload seamlessly. It maintained consistent performance during its most challenging scenarios: simultaneous data ingestion waves and query spikes that had crashed other solutions. Most importantly, Zilliz Cloud enabled sophisticated features, such as finding potential referrers within companies through queries that combine multiple vector searches with graph-like relationship analysis—capabilities that were previously impossible with other platforms.
Unlike self-hosted solutions, Zilliz Cloud required no database administration overhead, allowing the team to focus entirely on product development. The platform's dynamic schema support lets Jobright experiment with matching algorithms in real-time, deploying improvements during business hours without worrying about system stability.
The Role of Vector Infrastructure in Real-Time Matching
Jobright now uses six to seven specialized Zilliz Cloud instances, each optimized for a specific type of query:
Core Job Matching Engine: Matches user profiles against millions of listings, factoring in similarity, location, experience, visa status, and more.
Referral Discovery: Finds potential referrers based on education, past employers, and connections—using relationship-aware vector queries.
Company Insights: Surfaces qualitative insights (e.g., culture fit, hiring patterns) by indexing company profiles.
Career Trajectory Modeling: Recommends roles based on evolving skills and time-weighted embedding vectors.
Before all this content is ingested, indexed, and retrieved in Zilliz Cloud, all data is encoded into high-dimensional vector embeddings using Jobright's specialized fine-tuned models. The team employs different embedding models optimized for specific content types throughout the system—job descriptions utilize models trained on professional language, while company culture descriptions employ models optimized for organizational characteristics and values.
As Jobright's matching algorithms evolve based on user feedback, they frequently add new vector dimensions and modify filtering criteria. Zilliz Cloud's flexible schema support enables these changes without system downtime, allowing the team to deploy algorithm improvements during business hours—a capability that has proven invaluable for maintaining their competitive edge.
The platform integrates Zilliz Cloud with a comprehensive technology stack designed for scalability and reliability. Built on AWS with auto-scaling groups and load balancers to handle traffic spikes, Zilliz Cloud instances are distributed across availability zones for high availability. Integration with multiple job board APIs, company websites, government databases for H-1B data, and professional networks ensures comprehensive and up-to-date job listings flow seamlessly into the vector database infrastructure.
Jobright also leverages Zilliz Cloud's advanced features to enable sophisticated search scenarios that combine multiple types of similarity and filtering in ways that were previously impossible. Complex queries like "machine learning roles at startups offering visa sponsorship within 50 miles of San Francisco" combine semantic vector search with categorical filters and geographic constraints in a single operation. Finding suitable referrers requires searching across multiple vector spaces simultaneously—educational background embeddings, work experience vectors, and company relationship mappings. Career progression recommendations consider how user skills and interests evolve over time, using time-weighted vector operations to predict suitable next career moves.
From 500ms Timeouts to Instant Search—and Zero Admin Headaches
The transition to Zilliz Cloud delivered immediate and sustained improvements across all key metrics, transforming Jobright from a struggling startup into a leading industry player.
From 500ms Timeouts to 50ms Consistency
Query response times improved dramatically, from inconsistent 200-500ms with frequent timeouts to consistent sub-100ms performance, with a P95 latency of under 50ms for core matching operations. The system achieved 99.9%+ uptime after deployment, eliminating the frequent outages and performance degradations that had plagued their previous infrastructure. The platform seamlessly scaled from handling thousands of daily queries to processing over 2 million user interactions daily without requiring infrastructure changes or experiencing performance degradation.
Zero Database Administration Overhead
"We went about a year without having to manage this system at all. And I literally forgot we were using it," Ethan notes. "That's the highest compliment I can give infrastructure—when you stop noticing it exists, it's working perfectly."
This operational simplicity enabled the small engineering team to focus entirely on product innovation, rather than database administration, significantly accelerating their development velocity.
Beating LinkedIn and Indeed in User Engagement
These infrastructure improvements directly translated to superior user experiences and industry-leading engagement metrics. According to SimilarWeb analytics, Jobright now has the highest average user visit duration of any job search platform, surpassing LinkedIn, Indeed, and all other competitors. Daily active users surged from 10 in the first year to nearly 100,000, with users averaging over 40 platform interactions per session—significantly higher than industry benchmarks. Advanced features, such as referral discovery and AI-powered application assistance, achieve high adoption rates because they perform reliably and deliver results quickly.
Faster Feature Deployments Without Downtime
Reliable infrastructure enabled Jobright to maintain its competitive advantage through rapid feature development. The team now ships new features weekly with confidence that the underlying infrastructure won't break during updates or traffic spikes. Monthly retention rates improved significantly as users experienced consistent, fast performance and increasingly accurate job recommendations powered by continuously evolving algorithms.
Predictable Costs That Decrease Per User
The cost benefits proved equally transformative. Jobright moved from unpredictable costs associated with managing failing infrastructure to predictable, usage-based pricing that scales with business growth. Database administration overhead was eliminated entirely, and per-user infrastructure costs decreased as the platform grew, creating favorable unit economics that supported sustainable business expansion.
Looking Forward: Scaling the Vision
Jobright's vision extends far beyond job search to become a comprehensive talent marketplace connecting employers and candidates more efficiently than ever before. Advanced analytics will enable companies to understand talent market dynamics, competitive positioning, and optimal hiring strategies, utilizing the same vector infrastructure that powers candidate matching. Automated skill verification and assessment tools will use vector similarity to evaluate candidate capabilities against role requirements, while real-time labor market insights derived from the platform's comprehensive data will become valuable intelligence for employers and policymakers seeking to understand workforce trends.
Jobright also plans to expand to European and Asian markets, with Zilliz Cloud’s global infrastructure enabling low-latency service delivery across different continents. The flexible embedding infrastructure supports market-specific model deployment, allowing Jobright to adapt to local hiring practices, skill requirements, and career progressions without major system overhauls.
"We're planning international expansion while adding B2B services for employers," Ethan explains. "We need infrastructure that can handle multiple markets, languages, and regulatory requirements without rebuilding our core systems."
Lessons for AI Builders
Jobright's remarkable growth from 10 to 100,000 daily users offers valuable insights for AI entrepreneurs navigating similar challenges. Ethan generously shared several observations from their journey:
Proximity to users matters more than technology sophistication. The best AI in the world won't help if you don't understand your users' real problems. Jobright succeeds because they've invested more than anyone else in understanding job seekers' actual experiences through weekly user interviews and feedback loops.
Founders must stay hands-on longer than they expect. The temptation to hire specialists early is strong, but premature delegation often leads to failure. Founders must master critical capabilities, such as growth marketing and user research, before they can be delegated effectively.
The frequency of decision-making matters more than the speed of decision-making. While staying informed about rapid AI developments is crucial, making strategic pivots too frequently can destroy team momentum and erode user trust.
Infrastructure choices compound over time. Small technical decisions create significant competitive advantages at scale. The right infrastructure enables innovation rather than just solving immediate problems.
"When we first chose Zilliz Cloud, we thought we were solving a scaling problem," Ethan reflects. "But it actually solved an innovation problem. Having a reliable, powerful vector search infrastructure lets us experiment with matching algorithms we couldn't have attempted otherwise."
Ethan emphasizes that many AI startups underestimate the operational overhead of managing sophisticated infrastructure. In fact, infrastructure reliability directly impacts development velocity—unreliable systems slow feature development while reducing confidence in shipping new capabilities.
In markets where everyone has access to similar AI models, competitive advantage increasingly comes from infrastructure decisions that enable superior user experiences and faster innovation cycles. __
"Our strongest competitive moat isn't our AI models—it's our ability to deploy those models at scale with an exceptional user experience," Ethan concludes. "Zilliz Cloud gave us that capability."
Conclusion: Powering the Next Generation of AI Applications with Scalable Infrastructure
The partnership between Jobright and Zilliz Cloud shows how enterprise-grade infrastructure enables the development of breakthrough AI applications. Jobright focused on understanding users and building superior product experiences, while Zilliz Cloud provided the reliable, scalable foundation that made those experiences possible at scale.
As AI applications become increasingly sophisticated, the infrastructure layer becomes more critical to success. Vector databases aren't just a technical requirement—they're an innovation platform that determines what's possible for AI-powered products.
"We're proud to power Jobright's remarkable growth," notes James Luan, VP of Engineering at Zilliz. "Their platform demonstrates what's possible when innovative AI algorithms meet enterprise-grade infrastructure."
For Jobright, the infrastructure choice made one year ago continues to pay dividends as they prepare for the next phase of growth.
"Choosing Zilliz Cloud was one of our best early decisions," Ethan concludes. "It enabled us to build the product we envisioned rather than the product our infrastructure limitations would allow. In AI applications, that difference often determines success or failure."
- About Jobright: Redefining Job Search in the AI Era
- Why Traditional Search Broke at 2 Million Queries a Day
- How Jobright Benchmarked the Vector Database Market
- The Role of Vector Infrastructure in Real-Time Matching
- From 500ms Timeouts to Instant Search—and Zero Admin Headaches
- Looking Forward: Scaling the Vision
- Lessons for AI Builders
- Conclusion: Powering the Next Generation of AI Applications with Scalable Infrastructure
Content
Our strongest competitive moat isn't our AI models—it's our ability to deploy those models at scale with an exceptional user experience. Zilliz Cloud gave us that capability.
Ethan Zheng