Build a Custom Voice AI for Candidate Prequalification
Yes, custom voice AI tools can pre-qualify candidates for niche industries. These systems ask specific technical or compliance questions standard tools cannot handle.
The scope of a build depends on the number of screening questions and the API quality of your Applicant Tracking System (ATS). A system asking five yes/no questions that integrates with Greenhouse is a 2-week build. A system asking ten open-ended questions that must handle regional accents and connect to a legacy on-premise ATS is a 4-week build.
We built a voice AI pre-qualifier for a 15-person specialized recruiting firm placing certified welders. Their team spent hours each day calling applicants just to verify AWS D1.1 certifications. The system went live in 3 weeks, saving each recruiter 8 hours per week of manual screening calls.
What Problem Does This Solve?
Recruiting teams often try generic AI interview platforms first. These tools are designed for behavioral questions like “Tell me about a time you faced a challenge,” but they cannot verify hard skills. They use broad keyword matching that fails when asked to confirm a specific, technical detail like a “TWIC card” or “6G pipe welding certification.” The AI can't distinguish a real certification from a candidate just mentioning the term.
Next, teams try building a phone tree with an IVR service. This approach is too rigid. An IVR follows a strict script and cannot handle natural conversation. If a candidate says, “My D1.1 is active but my D1.5 is pending,” the system marks it as a failure because it expected a simple “yes.” Candidate drop-off rates for these systems often exceed 70% because the experience is frustrating.
Connecting a web form to a Twilio voice message seems clever, but it’s brittle and expensive. The transcription task for a 5-minute call can cost over a dollar, and the logic can't handle dropped calls or retries. You end up with a high API bill for a system that provides incomplete transcripts and loses qualified candidates.
How Does It Work?
We start by mapping your 5-7 most critical, non-negotiable pre-qualification questions. For a medical device recruiter, this included specific FDA compliance knowledge and experience with 510(k) submissions. We use this map to define the conversational flow and scoring rubric, which we store in a Supabase table. This allows you to update questions without a new code deployment.
We build the core logic as a Python service using FastAPI. This service manages the state of the conversation and interacts with the Claude API to understand candidate responses. The model doesn't just check for keywords; it interprets intent. A candidate saying “I’ve been running those submissions for five years” is correctly scored as meeting a “3+ years of experience” requirement. The audio processing pipeline achieves a P95 latency of under 400ms.
The FastAPI service is deployed on AWS Lambda, ensuring you only pay for compute time during active calls. When a candidate applies through your website, a webhook from your ATS (like Bullhorn or Lever) triggers the Lambda function. Once the call is complete, the system posts the transcript, a pass/fail score, and a 3-sentence summary back to the candidate record in your ATS within 90 seconds. We use httpx for all external API calls, with automatic retries for transient network errors.
We configure structured logging using structlog, sending all operational data to a central dashboard. You can monitor call volume, completion rates, and average scoring accuracy. If the pass rate for a specific job requisition drops by more than 20% in a 24-hour period, a Slack alert is sent to your team. Hosting and API costs for processing 500 candidates a month are typically under $60.
What Are the Key Benefits?
Get a Shortlist in Hours, Not Weeks
The system calls every new applicant automatically. Recruiters start their day with a pre-qualified list instead of a backlog of 200 resumes to screen manually.
One-Time Build, No Per-Seat License
After the initial fixed-price build, you only pay for raw cloud and API usage. There are no recurring license fees that increase as your team grows.
You Own the Source Code
We deliver the complete Python source code to your company's private GitHub repository. The system is a permanent asset, not a temporary rental.
Monitors Itself, Alerts on Failure
The system includes health checks and structured logging. If your ATS API is down, calls are automatically rescheduled. You get a Slack alert for any critical errors.
Integrates Natively With Your ATS
Call summaries and scores are written directly into candidate records in systems like Greenhouse, Bullhorn, or Lever. Your team works in one place.
What Does the Process Look Like?
Script and Scoping (Week 1)
You provide the top 5-10 knockout questions for a specific role and grant read-only API access to your ATS. We deliver a finalized call script and a technical integration plan.
Core AI Build (Week 2)
We construct the voice AI agent using FastAPI and the Claude API. You receive a secure endpoint to test the agent's conversational abilities with your own voice.
Integration and Deployment (Week 3)
We connect the voice agent to your ATS and deploy the full system on AWS Lambda. You receive a runbook detailing the architecture and maintenance procedures.
Live Monitoring and Handoff (Week 4)
We monitor the first 100 live candidate calls, tune the scoring logic based on real-world data, and then transfer full ownership of the GitHub repository.
Frequently Asked Questions
- How much does a custom voice AI pre-qualifier cost?
- A typical build is a 3-4 week engagement. The final cost depends on the complexity of the questions and the quality of your ATS API. A system that asks five yes/no questions and integrates with Greenhouse is straightforward. One that must understand industry jargon and connect to a legacy on-premise system requires more work. We provide a fixed-price quote after the discovery call.
- What happens if a candidate hangs up or the call drops?
- The system logs the incomplete call and automatically retries once after a 60-minute delay. If the second attempt fails, the candidate is flagged in your ATS for manual recruiter follow-up. This prevents qualified candidates from being lost due to technical issues. The system doesn't get frustrated and always follows the defined logic, unlike a human recruiter might after a bad connection.
- How is this different from a platform like HireVue?
- HireVue and similar tools are video-first platforms focused on assessing soft skills and behavioral traits. They are not designed to verify specific, technical, or compliance-related qualifications conversationally. Syntora builds a voice-first system engineered for hard-skill pre-qualification, asking the exact knockout questions your recruiters use to build their initial shortlist.
- Can the system handle different accents and dialects?
- Yes. We use modern transcription APIs trained on millions of hours of diverse audio, covering a wide range of accents. While no system is perfect, accuracy is typically above 95% for clear speakers. The Claude API is also robust enough to understand intent even with minor transcription errors, focusing on the core meaning of the candidate's response rather than just keywords.
- How do we change the screening questions after the system is live?
- The questions, scoring logic, and conversational flow are stored in a simple configuration file in a Supabase database, not hardcoded in Python. We provide documentation showing you how to edit these values directly. Changing a question is as simple as editing a row in a table. No new development engagement is required for routine updates to your screening criteria.
- How is candidate data handled for privacy and compliance?
- The system is built on your infrastructure, typically your own AWS account. Candidate audio and transcripts are processed in memory and are not stored long-term unless you explicitly configure it. We write only the final score and a non-PII summary to your ATS. This gives you full control over candidate data and helps with GDPR or CCPA compliance, unlike third-party SaaS tools.
Related Solutions
Ready to Automate Your Small Business Operations?
Book a call to discuss how we can implement ai automation for your small business business.
Book a Call