Build a Voice AI System to Prequalify Candidates
Choose a partner who builds a custom system you own, not one who rents you a SaaS platform. The right partner delivers full source code and integrates directly with your existing Applicant Tracking System.
The scope of a voice AI recruiter depends on the complexity of your questions and the API of your ATS. A 5-question screener that writes to a simple ATS is a two-week build. A system with dynamic question branching based on candidate answers and integration with a legacy platform requires more discovery.
We built a voice pre-screening agent for a 12-person recruiting firm that processes 400 applicants a month. The system went live in 2 weeks. It cut their time-to-initial-contact from 3 days to under 5 minutes and automatically screened out 60% of non-qualified applicants.
What Problem Does This Solve?
Recruiting teams often start with off-the-shelf voice AI platforms. These tools are rigid, designed for simple yes/no questions, and cannot handle the nuanced, multi-part answers required for real screening. They can ask "Do you have 3 years of experience?" but fail when a candidate says, "I have two and a half years at my current job and one year at my previous one." The system either fails or just saves a useless transcript, forcing a manual review.
A common failure scenario involves a nursing recruitment agency trying to pre-qualify candidates. Their key question is "Tell me about your nursing licenses, including state, number, and expiration date." A SaaS tool can only transcribe the audio. This leaves the recruiter to listen to 200 voicemails a week, defeating the purpose of automation. On top of that, per-call pricing of $0.75 adds up to $300/month for a glorified answering machine.
Some teams attempt a DIY solution using Twilio and a basic speech-to-text API. They quickly find that a raw transcript is not useful data. The real challenge is Natural Language Understanding: parsing the transcript, extracting structured entities like dates and license numbers, and scoring the answers. This is where DIY projects stall, burning engineering hours without producing a usable system.
How Does It Work?
We begin by mapping your 5-7 core pre-screening questions to a strict data schema using Python's Pydantic library. This defines the final JSON object we will post to your Applicant Tracking System (ATS), ensuring every candidate summary is structured identically. We then provision a dedicated phone number for your agent using the Twilio API.
The core of the system is a FastAPI application deployed on AWS Lambda for event-driven execution. When a candidate calls, Lambda invokes the application. We use the Claude API for the conversational engine. It understands the candidate's full response, asks clarifying questions if an answer is ambiguous, and maintains context throughout the 3-minute call. The latency for a complete conversational turn, from the end of the candidate speaking to the start of the AI's response, is under 800ms.
Once the call concludes, the system makes a final, asynchronous call to Claude using httpx to summarize the entire conversation and score it against your predefined rubric. This process generates a structured JSON payload with the candidate's answers, a 50-word summary, and an overall qualification score. The payload is then posted to your ATS's REST API. The entire post-call analysis and data entry completes in under 8 seconds. Transcripts and audio are stored in a Supabase Postgres database for 90-day retention.
For monitoring, we implement structured logging using structlog, sending all logs to AWS CloudWatch. We configure alerts that trigger if the API error rate exceeds 2% or if the average call processing time increases by 20%. This proactive monitoring ensures system health. Your monthly hosting costs for up to 1,000 candidates are typically under $25.
What Are the Key Benefits?
Live in 10 Business Days, Not 2 Quarters
Your custom voice agent starts screening candidates in two weeks, bypassing the lengthy sales and implementation cycles of enterprise SaaS tools.
Fixed-Price Build, Pennies Per Call
Pay a one-time fee for the build. Your ongoing costs are for the underlying API usage, not inflated per-seat or per-call SaaS fees.
You Own The Code and The Phone Number
We deliver the complete Python source code to your GitHub repo. You are not locked into a platform and can modify the system with any developer.
Writes Structured Data Directly to Your ATS
The system connects to Greenhouse, Lever, or any ATS with an API, populating custom fields with structured data, not just messy transcripts.
Alerts on Failure, Not Candidate Complaints
AWS CloudWatch monitoring alerts us instantly if a call fails to process, so we can fix it before your team or candidates even notice.
What Does the Process Look Like?
Week 1: Script and Schema Definition
You provide your screening questions and read-only API access to your ATS. We deliver a finalized conversational script and a Pydantic data schema for your approval.
Week 1: Core Application Build
We build the FastAPI application, integrate the Claude conversational engine, and configure the Twilio phone number. You receive a private number for internal testing.
Week 2: Integration and Deployment
We connect the voice agent to your ATS API and deploy the full system on AWS Lambda. You receive an invitation to the GitHub repository with all source code.
Weeks 3-4: Monitoring and Handoff
We monitor the first 100 live calls to tune prompts for accuracy. You receive a runbook detailing system architecture, monitoring dashboards, and operational procedures.
Frequently Asked Questions
- What factors determine the project's cost?
- The primary factors are integration complexity and question logic. Integrating with a modern ATS like Greenhouse is straightforward. Connecting to a custom, internal database takes more time. Similarly, a linear 5-question script is simpler than one with complex branching logic where the next question depends on the previous answer. We scope this during our discovery call.
- What happens if a call drops or a candidate hangs up mid-call?
- The system is designed for graceful failure. If a call ends prematurely, the partial transcript and extracted data are still processed. The system posts the incomplete record to your ATS and adds a specific tag, like 'Call Incomplete', signaling a recruiter to follow up manually. No data is lost.
- How is this better than using a BPO call center?
- A BPO uses human agents who are inconsistent, require constant training, and work limited hours. Our AI agent works 24/7, asks questions with 100% consistency, and costs a fraction of a human's hourly rate. It also delivers structured data directly to your systems, eliminating manual data entry from call notes.
- Can the AI handle different languages or strong accents?
- Yes. The underlying models from Anthropic and AWS support over 50 languages. We configure the agent for the specific languages your candidate pool uses. For strong regional accents, we can add examples to the system's prompt to provide more context, which significantly improves transcription and understanding accuracy without any custom model training.
- How do we update the screening questions after the build?
- The questions and scoring rubric are stored in a simple JSON configuration file within the codebase. Changing the wording of a question is a one-line text edit that a non-engineer can make through GitHub. Adding entirely new questions that require changes to the data schema would be a small, separate scoped project, typically a few days of work.
- Is the system compliant with privacy regulations like GDPR?
- Yes, because you control the infrastructure. The system is deployed in your own AWS account, making you the data controller. We configure data retention policies, such as auto-deleting call recordings after 30 days, to match your compliance requirements. Unlike a SaaS vendor, we never store or have access to your candidate data post-launch.
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