Jaspi
AI Recruitment Platform
Jaspi reimagines the hiring pipeline. Traditional recruitment is slow, biased, and expensive. Jaspi solves this by deploying AI agents that conduct structured interviews, evaluate responses in real-time using RAG against job specifications, and produce ranked, explainable candidate reports — all without human intervention.
Tech Stack
The Problem
Manual recruitment screening is slow, inconsistent, and fails to scale. Companies lose top candidates due to weeks-long processes, and HR teams spend 80% of their time on low-signal tasks like first-round interviews.
Architecture
Built a multi-agent system with LangChain where specialized agents handle different interview domains (technical, behavioral, domain-specific). A RAG pipeline indexes job descriptions and company culture docs in Pinecone. Each interview session creates a context window with retrieved relevant facts, enabling highly targeted questions.
Engineering Challenges
- ▸Maintaining conversational coherence across long multi-turn AI interviews
- ▸Preventing hallucinations in candidate evaluation reports
- ▸Real-time streaming responses with low latency at scale
- ▸Designing agent handoff protocols for specialized interview domains
Solution
Implemented a supervisor-agent architecture where a coordinator agent routes conversation segments to domain specialists. Used structured output parsing with Zod schemas to ensure evaluation reports are grounded in retrieved evidence. Redis pub/sub handles real-time streaming.
Results & Impact
- ✓Reduced time-to-screen from 2 weeks to under 24 hours
- ✓85% reduction in HR screening costs
- ✓Consistent, bias-reduced evaluations across all candidates
- ✓Successfully deployed for 3 enterprise clients