Get My Parking

Tech

Behind the Scenes: Building AVA, Our AI Voice Agent for ExpressLane

Jan 16, 2026

Behind the Scenes: Building AVA, Our AI Voice Agent for ExpressLane

Ever been stuck at a garage exit with cars honking behind you? That frustrating moment sparked the creation of AVA, our AI voice agent designed to solve parking’s most common headaches.

The Challenge

Parking operators handle thousands of support calls around the clock. Nearly 70% of those calls involve the same basic issues: declined cards, lost tickets, and gate malfunctions. Moreover, these repetitive calls drain time, resources, and budgets that could be better spent elsewhere.

The Solution: AVA

Our team built AVA as a voice-native AI agent integrated deeply into our ExpressLane platform. However, this wasn’t just about slapping AI onto an existing system. Instead, it was a journey of iteration, technical challenges, and thoughtful design decisions.

Here’s what made AVA special:

  • Real language understanding handles accents, slang, and incomplete sentences
  • Voice-first architecture delivers sub-minute resolution times across English, Spanish, and French
  • Integration with 2,800+ tools from our ExpressLane platform enables real-time decision-making
  • AVA achieves an 80% automation rate for high-frequency call types without human escalation

The Build Journey

We started by piecing together STT, LLM, and TTS models before pivoting to multimodal systems. Along the way, we ran hundreds of prompt iterations to find the perfect voice personality. Furthermore, we built hardware kiosks with noise-reducing mics and crafted handoff logic for human escalation.

The process involved testing multiple AI models (OpenAI, Google Gemini, and others). Simultaneously, we built all-weather kiosk hardware from a software-first foundation. We also treated prompts like code, versioning, testing, and deploying them with discipline. Finally, we created a personality that’s empathetic yet efficient, warm yet authoritative.

What We Learned

Prompts need to be treated like code. Hundreds of iterations were necessary to remove ambiguity and lock down precise actions.

Integration is everything. AVA’s power comes from deep platform integration. Consequently, it accesses live transaction data, payment status, camera feeds, gate controls, and more to eliminate hallucinations.

Strategic positioning matters. AVA isn’t a standalone product. Rather, it’s an embedded feature of ExpressLane. Therefore, we designed it to handle the repetitive 70% and seamlessly escalate complex cases to human experts.

The Result

An AI agent that feels like a parking concierge, not a bot. It’s focused, empathetic, and built to get parkers out quickly while keeping lanes clear for operators.

Want the full story? Rajesh Prabhu, our CTO, shares the complete technical journey. In detail, he covers the challenges we faced, our partnership with Google GCP, pilot results from New Jersey and Georgia, and what’s next for AVA’s 2026 launch.

👉 Read the full article on LinkedIn to dive deep into how we built AVA from the ground up.

Curious about AI in parking? Have questions about voice agents? Drop us a comment or reach out. We’d love to hear your thoughts!