From: aidotengineer

Building successful AI applications, particularly in personal health, requires a strong focus on user trust and community engagement [08:26:00]. Alma, an AI nutrition companion, has identified several key pillars in achieving this.

Core Principles for an AI Nutrition Companion

Alma’s vision for an AI nutrition companion is built on three core pillars [00:00:58]:

  1. Simplified Nutrition Tracking: Making tracking natural and easy, akin to texting a friend, rather than navigating complex lists or relying on imprecise photo recognition [00:01:04].
  2. Contextual Understanding: Utilizing tracked information to build a deep understanding of the user’s flavor profile, interests, habits, and hobbies to guide them towards their goals [00:01:23].
  3. Goal-Oriented Recommendations: Connecting users with relevant products, restaurants, and meals that help them achieve their nutritional objectives [00:01:36].

Alma also introduced the “Alma Score,” a score out of 100 developed with Harvard University, to help users easily understand the holistic quality of their food choices and encourage healthier eating habits [00:02:31].

Strategies for Building Effective AI Agents and Trust

Prioritizing User Feedback

While evaluation metrics are important, nothing surpasses real-time, in-the-moment user feedback for guiding AI development [00:03:12]. Alma implemented a “How did Alma do?” prompt after every interaction to gather direct input, which has been crucial for measuring accuracy and driving improvements [00:03:18].

Constraining LLMs for Accuracy

Large Language Models (LLMs) given open-ended tasks often result in high error rates [00:03:41]. It’s vital to constrain the LLM to very specific tasks to improve accuracy [00:03:48]. Alma achieves this by breaking down complex processes, like tracking a food item, into smaller, sequential steps [00:04:06]. For example, recognizing “banana,” extracting it as a food item, and then processing it with a USDA database for caloric content, all in distinct steps, makes the user experience feel faster and more reliable [00:04:21].

Understanding User Value beyond Core Functionality

Features that might seem minor can hold significant value for users. Alma discovered the importance of “streaks” when a bug broke them, leading to an outcry from the user community [00:05:06]. This highlighted that users deeply value such engagement elements, prompting Alma to double down on them [00:05:25].

Continuous Context Building

To prevent user frustration from repetitive input, AI agents must continuously build context around the user [00:05:41]. Alma’s “About You” feature automatically adds novel pieces of information about the user to a knowledge dataset, allowing Alma to learn and get smarter with every interaction [00:05:53]. Users can also view, delete, or add to this information themselves [00:06:08].

Proactive Insights and Education

Engagement can be enhanced by proactive insights from the AI [00:06:24]. Alma detects insights about users’ eating habits and surfaces information they might not know, fostering a continuous learning process in a bite-sized, engaging format [00:06:32].

Offering Multimodality

Instead of being overly focused on a single interaction modality (e.g., voice, text, photo), providing users with multiple options increases usability and satisfaction [00:07:07]. Users appreciate the flexibility to choose the most convenient method for their context [00:07:31].

Future Focus Areas for Building Trust and Community

Alma is doubling down on three key learnings for the future of AI applications [00:07:54]:

  1. Brand and Design: Code is becoming commoditized, making user-centric design and visual appeal critical for standing out [00:07:59].
  2. Trust and Partnerships: To earn trust, especially as a new entity, it’s important to partner with trusted sources and individuals who have an aligned mission [00:08:26]. Alma actively asks users where they get information and seeks partnerships accordingly [00:08:36].
  3. Community and Net New Data: As AI models “hoover up” vast amounts of online data, the focus shifts to leveraging user communities to create net new data that is valuable and interesting [00:08:48]. Users are curious about how others eat, which fosters community and generates unique insights [00:08:57].