Traba deploys AI interview agents to scale industrial staffing

Traba is building a next-generation staffing platform for the industrial supply chain. Their mission is to match businesses with vetted, qualified temp workers at scale.

To reach that goal, Traba built Scout, an AI-powered interviewing system integrated directly into their operations. Scout now conducts over 50,000 interviews monthly across warehousing, logistics, and manufacturing roles - reducing manual workload, improving placement rates, and delivering consistent evaluations across every region.

Staffing is the bottleneck in industrial supply chains

While millions of workers are ready to work, high-friction hiring processes hold back fulfillment centers, logistics hubs, and manufacturers from operating at peak efficiency.

These jobs require qualification. Shift schedules vary. Language barriers exist. Regulatory requirements must be followed. All of this slows down staffing.

Traba needed to scale without hiring thousands of recruiters. They needed a consistent, reliable system that could assess worker fit faster.

Why Traba chose ElevenLabs

In late 2024, real-time Text to Speech and Speech to Text became viable for phone-based interviewing. Traba began testing vendors with one goal: find a partner that could support advanced conversational AI without requiring full pipeline ownership.

ElevenLabs offered:

• High-quality voices: natural, multilingual voices that made conversations feel human rather than robotic. 
• Low latency: fast enough for real-time interactions without awkward delays. 
• Flexibility and control: the ability to orchestrate multiple agents, experiment with prompting strategies, and integrate directly into their systems.
• Reduced complexity: handling challenging parts of the audio pipeline so they can focus on their unique workflows.

Building the AI interviewer

Scout launched with a single-agent architecture. Its first version proved that AI could conduct structured interviews, qualify candidates, and return useful evaluations.

Scout V1:

• Monolingual: Only supported English, limiting reach
• Single-agent logic: One LLM handled all steps — introduction, Q&A, logistics
• Static question sets: Role-based, predefined queries with limited flexibility
• Basic evaluation: One-pass summary prompt at the end of the interview
• Operator handoff: AI provided directional signal; humans made final decisions

Despite its simplicity, V1 ran thousands of calls in parallel and delivered time savings immediately.

Scaling to 250,000+ calls: solving for depth, speed, and consistency

By March 2025, Scout had run over 17,000 interviews and saved more than 1,400 hours of manual vetting time. To prepare for peak seasonal demand, the system was rebuilt to operate autonomously.

Key upgrades included:

Multilingual voices and dynamic switching

ElevenLabs shipped multilingual support, enabling Scout to switch between English and Spanish mid-call based on user preference. This unlocked access to a previously underserved worker segment.

Multi-agent orchestration

As the interview context expanded, Traba encountered model degradation. ElevenLabs provided the tools to split calls across specialized agents - introduction, vetting, logistics, and FAQ support - with seamless transitions during the conversation.

Deduplicated interview logic

Workers applying for multiple jobs were getting asked the same questions. Traba engineered a preprocessing pipeline to deduplicate semantically similar questions across interviews. This reduced redundancy by up to 20% per candidate.

Custom evaluation framework

Operators needed more control over assessments. Traba built Custom Scout, a framework to define what ‘good’ answers look like on a per-question basis. Evaluations now align with each client’s unique criteria.

Ground truth feedback and prompt iteration

Traba developed an internal prompt testing framework with instant feedback loops. By generating human-verified datasets through Langfuse, the team could A/B test prompts against real-world performance — allowing fast iteration at scale.

Results

Traba’s AI-led interviewing system now powers over 50,000 interviews per month and 85% of all worker vetting across the platform is fully automated. At an average of 5 minutes per conversation, this saves over 4k operator hours per month.

• 15% higher shift completion rates for AI-qualified workers vs. human-qualified
• Consistent assessments across roles, shifts, and geographies
• Reduced time-to-hire with structured decision-grade evaluations
• Scalable vetting layer that runs 24/7 with minimal operator input

By refining their question banks, evaluation logic, and call flows through continuous feedback, Traba built a system that scales while improving outcome quality.

What’s next

Traba’s roadmap includes agent-led onboarding, video-based Q&A, timesheet processing, and emotion detection via multimodal LLMs. They’re also working on agent-led prompt refinement, using performance data to train agents that optimize interview design autonomously.

Throughout this journey, Traba continues to partner with us as we develop the next generation of our Agents Platform, pushing the boundaries of language intelligence in complex real-world workflows.