The MIT Startup Exchange supports MIT-connected ventures as they explore and assess new technologies. Here are 10 AI-focused companies from the Startup Exchange that were featured at the 2025 MIT AI Conference in April.
Market pain point: Bridge collapses are mainly caused by poor maintenance. Visual inspections, the main assessment tool for bridges, are subjective, costly, and infrequent. Existing data-based monitoring solutions lack actionable insights, which management companies need.
Solution: Displaid is a bridge monitoring service that uses wireless sensors to collect data, and an AI algorithm to analyze the data and suggest actionable insights.
How it works: Displaid’s bridge monitoring technology uses AI trained on years of data from more than two dozen bridges to identify potential problems. For example, Displaid helped prevent a bridge closure in Italy, saving the Italian government 300,000 euros. Its sensors are 70% cheaper and three times more efficient than existing measures in identifying early-stage structural anomalies.
Market pain point: Computation demand is increasing at a faster rate than graphics processing units are improving.
Solution: Eva is developing a digital twin platform that will shorten AI model training times, leading to lower costs for enterprises that need high-compute capabilities.
How it works: Eva uses semiconductor technology and hardware-software codevelopment for compatibility and ease of deployment. Its digital twin platform has 72-times the throughput per dollar of the Nvidia Blackwell chip, drastically reducing Llama 3.1 model training time from 80 days to less than two days and the cost from $47 million to $500,000.
Market pain point: Forestry involves biodiversity management and wildfire risk mitigation. Current data collection methods, like using calipers and tape measures, are time-consuming, costly, and often inaccurate.
Solution: Gaia AI uses advanced algorithms specifically tailored to forest environments to measure and map trees 100 times faster than current methods. This below-the-canopy data is then used to train AI algorithms for analyzing satellite imagery so that the information can be scaled to much broader forested areas.
How it works: Gaia AI uses a lightweight backpack equipped with lidar for data capture, and mobile and web apps for navigation and data analytics. This year, Gaia AI is working with the U.S. Forest Service to scale wildfire risk management across the country.
Market pain point: Sales teams pursue many leads that don’t convert to sales. Long sales cycles and large top-of-the-funnel volumes result in low conversion rates despite massive effort. The average length of the cycle for selling technology into a health care organization is 19 months.
Solution: GetIntro creates a narrower top of the sales funnel for the health care industry by using an AI agent that introduces sellers to prospects that are both receptive and reachable.
How it works: GetIntro uses AI trained on conversations with hundreds of health care sales executives to find and facilitate introductions to potential customers and partners within the user’s extended network. GetIntro is working with Venture Lane, a B2B startup hub with a network of over 250,000 people.
Market pain point: Businesses rely on software, but the different systems they use don’t always work together. This results in disjointed workflows that employees must reconcile manually, which is costly and inefficient.
Solution: Perygee acts as the glue for all data sources and workflows within an enterprise, providing automated insights and actions across enterprise operations.
How it works: Perygee’s software integrates siloed systems, automating manual work. More than 25 companies, including PepsiCo and Burger King, are using it. Logistics company iGPS used Perygee’s AI model to automate its paper-based delivery confirmation process by detecting new forms, validating them, and flagging suspicious forms for customer review. IGPS has been able to save 80 hours of work each month, improve cash flow by two days, and elevate customer experience.
Entrepreneurship Development Program
In person at MIT Sloan
Register Now
Market pain point: Trucking and transportation delivery companies face challenges with drivers, including high turnover, subpar performance, and accidents.
Solution: SafeMode has developed a behavioral modification program with the goal of improving driver behavior for safer, more efficient, and more sustainable driving. Incentive programs help drivers earn more money based on strong performance.
How it works: SafeMode uses AI models to analyze data from existing fleet vehicle devices, score driver behavior, and determine cash incentives through a gamified app. Using SafeMode resulted in a 71% decrease in safety violations for Amazon Delivery Service Partners and a 30% decrease in accidents and insurance claims at Pinch Transport.
Related Articles
Market pain point: Data privacy concerns are hindering enterprise AI adoption, since companies are often required to share sensitive data with AI applications. Current data privacy solutions are inadequate for AI systems due to enforcement difficulties and incompatible cybersecurity applications. As a result, many enterprises are shifting to on-premises AI deployments, despite the large amount of engineering effort that entails.
Solution: Tinfoil provides a way to securely move confidential AI-powered applications into the cloud. Data is encrypted from the source all the way through the inference process, so even Tinfoil cannot see a customer’s data or models. Its infrastructure is transparent and verifiable, meaning that users can check to ensure that confidentiality is being maintained.
How it works: Tinfoil builds on top of Nvidia graphics processing units and employs confidential computing, creating a secure enclave on a server that receives and processes the encrypted data before sending it back to the user. Tinfoil can be easily integrated into existing AI deployments.
Market pain point: Manufacturing plants often lack sufficient data, especially regarding human factors like operator deviations. Self-reporting is the only way to get this information, but it can be unreliable.
Solution: Tristar AI has developed a vision system to inspect the production lines at manufacturing plants. Any deviations in standard operating procedure are flagged, and the operator is instantly alerted so that the problem can be fixed in real time.
How it works: Tristar AI uses modular “Lego-like” movement blocks to analyze complex motions in production processes, reducing the scrap rate by over 60% in one month and saving thousands in lost production and rework for clients. This technology provides real-time visibility to engineers and operators, enabling them to address issues proactively and making manufacturing smarter, leaner, more agile, and more resilient.
Market pain point: Large language models are trained on internet text and don’t consider human behavior, which leads them to focus on text features instead of actions. Behavior-driven businesses, like retail, rely on transactional and behavioral data, not text-based intelligence, to build competitive advantage. For example, an LLM may assert that lactose-free milk and regular milk are similar products, but consumers don’t usually buy them together. Taking consumer behavior into consideration, lactose-free milk and gluten-free bread are more closely related products.
Solution: Unbox AI has created its own LLM, called BehaviorGPT, to predict consumer behavior.
How it works: BehaviorGPT has been trained on over 1 trillion human actions to help organizations anticipate and respond to consumer behaviors. On average, Unbox AI’s BehaviorGPT has helped increase sales for its users by 20%. This is due in part to businesses’ ability to build new revenue streams based on BehaviorGPT recommendations.
Market pain point: Workforce fatigue in industries like transportation, mining, construction, and aviation leads to over $140 billion in economic losses yearly. Current fatigue management solutions rely on self-reporting or reactive monitoring, which are ineffective.
Solution: Vocadian is a predictive fatigue risk management system that informs managers on how to optimize schedules and personalize interventions.
How it works: Companies input their demographics and baseline vocal recordings from employees. Vocadian’s technology uses an AI algorithm based on voice biomarkers and circadian science to analyze each worker’s audio submission for fatigue risk. Early tests in Latin America and the U.S. have shown a 26% decrease in risk incidents and a 92% increase in productivity.
Browse the slideshow on Instagram
Kristina DeMichele
Assistant Director, Digital Marketing
[email protected]
