Build a Custom Voice AI Recruiting Screener
The best voice AI recruiting platform for an SMB is a custom system built on a large language model. Off-the-shelf tools have rigid scoring and high per-seat costs that small teams cannot justify.
A custom build is tailored to your specific roles and screening questions. It integrates directly with your existing Applicant Tracking System (ATS), providing recruiters with scored summaries without changing their workflow. The scope is defined by the complexity of your ATS and the number of unique roles to screen for.
We built a voice screening system for a 12-person recruiting firm processing 400 applicants/month. Their team previously spent 10-15 minutes per candidate on manual phone screens. The AI screener now handles the initial screen in 90 seconds and delivers a transcript with a scored summary, saving them over 60 hours per month.
What Problem Does This Solve?
Most recruiting teams try SaaS tools like MyInterview or Willo for initial screening. These platforms fail because their scoring logic relies on simple keyword matching. A candidate who mentions 'Python' gets points, but the system cannot distinguish between ten years of production experience and a three-month bootcamp. The scoring is too shallow to be reliable.
This forces recruiters to listen to every single audio recording anyway, which defeats the purpose of the tool. They end up paying a $499/month subscription for a system that creates more work. The core issue is that these tools cannot conduct a dynamic conversation or understand the nuance in a candidate's answer. They are a collection of recorded questions, not an intelligent interviewer.
Larger ATS platforms like Greenhouse offer AI add-ons, but these are bundled into enterprise tiers costing upwards of $25,000 annually. For a small or medium-sized business, this is not a viable option. You are forced to choose between a cheap tool that does not work or an expensive platform with features you will never use.
How Does It Work?
We start by analyzing your 3-5 most critical screening questions for a specific role. We use the Claude API to refine these questions and generate logical follow-up prompts. The entire system is a Python application built with FastAPI, which we deploy on AWS Lambda for efficient, consumption-based execution.
When a candidate calls the dedicated Twilio number, a webhook triggers the Lambda function. The system records their audio, transcribes it in under 3 seconds using a fine-tuned speech-to-text model, and sends the transcript to the Claude API. The prompt is engineered with your role requirements and a detailed scoring rubric. Claude scores the answers from 1-10 across 5 key criteria we define together, such as technical depth or communication clarity, and provides a written justification for each score.
The final output is a JSON object containing the full transcript, granular scores, and a 150-word summary of the candidate's fitness. This object is posted directly to your Applicant Tracking System as a note on the candidate profile. We use httpx to build direct API integrations for platforms like Greenhouse, Lever, or Ashby. The entire process, from the end of the call to the ATS update, completes in under 30 seconds.
We use Supabase to store a log of every screening for accuracy analysis and future prompt tuning. All logging is handled by structlog for machine-readable output, with alerts configured to post to a dedicated Slack channel if the transcription confidence drops below 90% or the Claude API returns an error. This infrastructure typically costs under $50 per month for up to 1,000 candidate screenings.
What Are the Key Benefits?
From Screen to Summary in 30 Seconds
Eliminate the lag between a phone screen and recruiter review. Get scored summaries in your ATS before the next candidate even calls in.
Pay for Usage, Not for Seats
A single fixed-price build with hosting costs under $50/month. No recurring per-recruiter SaaS fees that punish you for growing your team.
You Own the Scoring Logic and Code
The entire Python codebase is delivered to your GitHub. You can modify the screening questions and scoring rubric without asking for permission.
Alerts for Errors, Not for Invoices
We use structlog and Slack webhooks to send real-time alerts if transcription fails or an API key expires. You hear about problems, not billing cycles.
Works Inside Your Existing ATS
Direct API integration posts results into Greenhouse, Lever, or Ashby. Your recruiters stay in one system instead of toggling between multiple tabs.
What Does the Process Look Like?
Week 1: Scoping and Question Design
You provide access to your ATS and share your top 3-5 screening questions for one role. We analyze them and design the scoring rubric and follow-up logic.
Week 2: Core System Build
We build the core FastAPI application, connect to the transcription and language models, and set up the Twilio phone number. You receive a test number to try it yourself.
Week 3: ATS Integration and Deployment
We connect the system to your ATS API and deploy the application to AWS Lambda. You receive the full source code in your GitHub repository.
Weeks 4-6: Monitoring and Handoff
We monitor the first 100-200 live screenings, fine-tuning prompts for accuracy. You receive a runbook detailing how to change questions and manage API keys.
Frequently Asked Questions
- How much does a custom voice AI screener cost?
- A standard build takes 3 weeks. Pricing is fixed-scope and depends on your ATS integration complexity and the number of roles needing distinct question sets. The build is a one-time cost, followed by an optional flat monthly maintenance fee. Book a discovery call at cal.com/syntora/discover for a detailed quote based on your specific needs.
- What happens if the AI misunderstands a candidate or the transcription is bad?
- The system logs transcription confidence scores. If a score is low, it flags the screening for manual review in your ATS. The full audio recording is always attached to the candidate profile, so recruiters can listen themselves in edge cases. This ensures you never miss a good candidate due to a technical glitch.
- How is this different from using an AI meeting notes tool like Fathom or Fireflies?
- Meeting note-takers are designed for passive summarization. They cannot conduct a one-way, structured interview or score responses against a predefined rubric. Our system is an active agent that guides the candidate through specific questions and performs a quantitative evaluation, not just a qualitative summary of a conversation.
- Can we use this for different roles with different questions?
- Yes. The system uses the job ID from your ATS to automatically select the correct script when a candidate calls. We can add a new role with a unique question set and scoring rubric in about 2-3 days. There is no technical limit to the number of roles you can support with the system.
- Where is the candidate audio and transcript data stored?
- All data is processed and stored within your own cloud infrastructure, typically AWS, and your Supabase database. Syntora does not store any Personally Identifiable Information on our systems after the project handoff. You have full control over the data, retention policies, and access logs, which is critical for compliance.
- What is the candidate experience like? Do they know they're talking to an AI?
- The system introduces itself as an automated screening assistant for your company. We find transparency is key. The experience is like leaving a structured voicemail, but with a few distinct questions. Since it's a dedicated line, there are no scheduling conflicts, and candidates can complete the screen 24/7.
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