By Dachi Choladze
An industry-wide research and development update from the Impel R&D team
When consumers think about buying a car, it’s more than just a transaction. For many, it’s a pivotal moment filled with excitement, deliberation, and trust. Customers share critical personal and financial details, trusting in the systems that support these experiences. It’s a highly considered, personal process that’s too important to rely on generic solutions that jeopardize security or accuracy. That’s why, at Impel, we’ve set out to redefine the standard for automotive AI, blending safety with best-in-class performance.
Through relentless innovation, Impel’s team of engineers and developers based in Tbilisi, Georgia has achieved two significant milestones. First, Meta showcased the success of our domain-tuned LLM completed in partnership with AWS Sagemaker at their highly regarded LlamaCon Conference.
Next, we released Archias, a cutting-edge domain-specific expert model built to improve AI safety in critical industries like automotive. Archias represents a major step forward in creating safer, more reliable AI systems. By focusing on five key threat areas—such as in-domain misuse, adversarial prompts, and pricing manipulation—it addresses real-world risks faced by dealerships and OEMs. With its transparent and collaborative framework for evaluation and learning, Archias sets a new standard for aligning AI with industry needs and safety priorities.
Better Performance with Domain-Tuned AI
Rapid growth in the adoption of Impel’s AI Operating System pushed our engineering and R&D team to think outside the box. How could we achieve model customization AND cost optimization at scale? The team replaced Impel’s third-party LLM with a fine-tuned version of Meta’s Llama model, integrated into our proprietary blend of automotive-specific models and foundational LLMs, deployed on AWS Sagemaker. This upgrade boosted model output accuracy by 20%, enhanced security, and significantly reduced operational costs, creating greater value for both dealers and their customers. Our verticalized large language model (LLM), optimized in collaboration with AWS, received recognition for its measurable business impact at Meta’s LlamaCon.
This wasn’t just theory in action; we delivered measurable results in real-world scenarios. From sales to service workflows, our AI doesn’t simply respond; it understands the automotive retail environment deeply and delivers successful dealer and OEM outcomes.
Democratizing safer AI for automotive with the Archais model
On the research side, Archias reflects our approach to accelerating the understanding and improvement of AI safety in automotive and other specialized domains.
This domain-specific model ensures advanced protection against prompt injections, adversarial queries, and other emerging threats. Why does this matter? Because as generative AI for automotive becomes a core part of customer-facing workflows, the stakes are exponentially higher. Errors or misuse in high-stakes transactions like automotive sales can pose? Significant financial and reputational risk to businesses.
In benchmark testing, Archias improved model output accuracy and performance by up to 20.7%.
By publicly releasing our dataset and evaluation framework on platforms like arXiv, we’re enabling the industry to collaborate and improve AI safety standards across the board. Together, we can create systems that protect against out-of-domain requests and malicious attacks while deeply understanding the unique workflows they serve.
Why verticalized AI matters
The automotive industry is a unique ecosystem. From online vehicle browsing to service management, workflows are complex, intricate, and layered. General-purpose AI, no matter how advanced, can’t deliver the adaptability, safeguards, or industry-specific outcomes required in these environments.
Impel’s AI Operating System is purpose-built to serve every dealership department—from marketing to merchandising, and sales to service. With over 33 billion consumer interactions and $8 billion in influenced revenue, we’ve demonstrated that aligning AI to the specific needs of an industry creates unparalleled value.
Looking to the future
Innovation doesn’t happen in isolation. It takes collaboration, transparency, and a deep understanding of who we serve. At Impel, this philosophy drives everything we do, from partnerships with AWS and Meta to publishing research breakthroughs like Archias.
But here’s my challenge for the industry: With so much at stake, who are you trusting with your business? Automotive dealerships operate in a high-stakes environment, balancing financial, operational, and reputational risks every day. Partnering with vendors isn’t a choice businesses can afford to take lightly. You need expertise. You need systems purpose-built to meet your unique needs. You need safeguards that protect your data, your customers, and your future.
At Impel, we’re setting the standard. But we’re also making it easier for others to follow by leading with transparency and sharing our research openly. The future of AI in automotive is one we’re building together, and we’re just getting started.