Building the AI That Rewrites the Hiring Process
Universal Perk designed and built Creva AI's full-stack recruitment platform from the ground up — AI resume screening, automated voice interviews powered by ElevenLabs and Deepgram, LangChain-orchestrated scoring pipelines, and ATS integrations that cut time-to-hire in half.
The Client
Who Is Creva AI
Creva AI is a recruitment automation platform built on the belief that the hiring process is fundamentally broken — too slow, too manual, and too inconsistent to serve either employers or candidates well. Their vision: automate every repetitive step in the early hiring funnel so recruiters can spend 100% of their time on what actually requires human judgment — relationships, negotiation, and final decisions. Universal Perk was brought in to build that vision from scratch.
Our Process
How We Built It
We started with deep discovery to understand the recruiter workflow end to end before writing a single line of code. The result was a focused, phased build that shipped an AI-capable MVP in four weeks.
Recruiter Workflow Audit
We shadowed recruiters across multiple roles to map every manual step in the screening and interview pipeline — identifying where AI could add the most leverage.
AI Architecture Design
Selected the right models for each job: OpenAI for reasoning, ElevenLabs for voice synthesis, Deepgram for transcription, LangGraph for pipeline orchestration.
ATS Integration Mapping
Audited target ATS platforms (Greenhouse, HubSpot) and designed a webhook-based integration layer that syncs candidate data without manual exports.
Build, Train & Evaluate
Built the full-stack platform and scoring system with LangSmith tracing live from day one — giving us complete observability over every model decision.
Deploy & Optimize
Launched to production with live monitoring, A/B testing on interview question sets, and weekly model performance reviews for the first 60 days.
Discovery
Pain Points We Uncovered
Before we could build the right system, we had to understand exactly where the recruiting process was breaking down.
Screening Bottleneck
Recruiters were spending 60–70% of their time manually reviewing resumes that didn't match role requirements — time that should have gone to interviewing and closing top candidates.
Scheduling Overhead
Coordinating first-round interviews across time zones required back-and-forth emails that delayed the pipeline by days, causing candidates to drop out before ever speaking to the team.
Inconsistent Evaluation
Without a standardized screening process, candidate quality varied by reviewer. Different interviewers asked different questions, making cross-candidate comparisons unreliable.
Fragmented ATS Data
Candidate data lived in multiple disconnected systems — job boards, spreadsheets, and an ATS that didn't talk to the rest of the stack. Nothing gave a single view of pipeline health.
Platform Scope
What We Built
Four core capabilities that automate every manual step in early-stage recruiting.
AI Resume Screening
An AI engine that evaluates applications based on actual job criteria — not keyword matching — and surfaces the most qualified candidates automatically.
AI Voice Interviews
A conversational AI system that conducts structured first-round interviews 24/7, asks consistent questions, and scores candidates on the spot — no scheduler needed.
Candidate Scoring & Ranking
A quantified fit score for every candidate, based on resume analysis, interview performance, and role requirements — giving recruiters a ranked shortlist in seconds.
ATS Integration
Native connectors to Greenhouse, HubSpot, and custom ATS platforms via webhook API — so candidate data flows automatically without manual entry or duplicate records.
Built With
The AI Stack We Used
Best-in-class models for every layer of the pipeline — voice, language, orchestration, and observability.
GPT-4o for resume parsing, question generation & semantic scoring
Hyper-realistic AI voice synthesis for interview delivery
Real-time speech-to-text transcription of candidate responses
Foundation model layer for cost-efficient inference at scale
LLM orchestration, prompt chaining & retrieval-augmented generation
Stateful multi-agent workflows for the full screening pipeline
LLM observability, tracing & performance monitoring in production
Infrastructure & Frontend
Results
What Creva AI Shipped
Measurable outcomes from day one in production.
70% Reduction in Screening Time
AI resume evaluation cut manual review from hours to minutes, freeing recruiters to focus on high-intent candidates instead of filtering inboxes.
2× Faster Time-to-Hire
Automated voice interviews eliminated scheduling delays entirely. Candidates complete first-round screening within 24 hours of applying, at any time.
78% Candidate Satisfaction Rate
78% of candidates preferred the AI-led early screening over traditional phone screens — citing speed, flexibility, and lack of scheduling friction.
Consistent Evaluation Across Every Role
Every candidate is assessed against the same structured criteria, removing reviewer bias and making shortlists genuinely comparable.
Live ATS Integrations
Greenhouse and HubSpot connected out of the box. Candidate data syncs in real time — no manual exports, no spreadsheet updates.
Pipeline Analytics Dashboard
Time-to-hire, cost-per-hire, and funnel conversion tracked live across every open role — giving leadership data they could actually act on.
Let's Talk
Ready to build your AI platform?
Whether you're building from scratch or adding AI to an existing product — we'll get your first system live within four weeks.