From: aidotengineer
Alma’s core belief is that eating well shouldn’t be difficult, and good health stems from simple, personalized nutrition [00:00:10]. Historically, understanding one’s eating habits has been challenging despite numerous apps available [00:00:17]. The problem identified is an imbalance: users are required to share extensive information but receive little meaningful feedback [00:00:30]. Alma believes that AI can solve this, creating a unique “nutrition companion” [00:00:42].
Alma’s Vision for AI Nutrition
Alma’s vision for an AI nutrition companion rests on three core pillars [00:00:58]:
- Simple Nutrition Tracking
- The goal is to make nutrition tracking easy, natural, and feel like texting a friend [00:01:04]. This avoids the laborious task of searching through endless product lists or relying on inaccurate photo recognition [00:01:13].
- Building User Context
- Once tracking is easy, the information gathered is used to build a strong context around the user, including their flavor profile, interests, habits, and hobbies [00:01:23]. This context helps steer them toward their stated goals [00:01:33].
- Connecting Users with Goals
- In the second half of the year, Alma plans to use this information to connect users with specific products, restaurants, and meals that will help them achieve their nutritional goals [00:01:38].
Key Learnings in Building the AI Nutrition Companion
Alma gained several key insights during its beta phase and development:
- Importance of User Feedback: Real-time, in-the-moment feedback from users is paramount [00:03:12]. Alma implemented a “How did Alma do?” prompt after every interaction to gather this crucial data [00:03:18].
- Constraining Large Language Models (LLMs): Giving LLMs open-ended, very large tasks results in high error rates [00:03:39]. It’s vital to constrain the LLM to very specific tasks [00:03:48]. Alma addresses this by breaking down information processing into multiple steps, sending partial results to the user as they become available. For example, when a user tracks a banana, Alma first recognizes the food item, extracts it, sends it to the app, and then processes further details like caloric content from a USDA database in subsequent steps [00:04:08]. This makes the experience feel faster and more interactive for users [00:04:46].
- User Value for Features: Even seemingly minor features, like “streaks,” can be highly valued by users [00:05:03]. When a streak bug occurred, users actively voiced their concern, highlighting the importance of doubling down on features that users find engaging [00:05:16].
- Building Continuous Context: AI agents benefit significantly from continuously building context about the user [00:05:39]. Alma incorporates a “About You” knowledge data set that is updated with novel information from user interactions, making Alma smarter over time and reducing the need for users to reiterate information [00:05:53].
- Proactive Engagement: Proactive outreach is crucial because users only open the app to ask questions so often [00:06:22]. Alma detects insights about user eating habits and proactively surfaces unknown information, such as the increased vitamin C absorption when blueberries are paired with dark chocolate [00:06:30].
- Multimodality: While some modalities (like voice) are personally preferred for their efficiency in tracking meals [00:07:09], users appreciate the option to use voice, photo, or text depending on the context [00:07:31]. Providing multiple modalities that make sense for users is essential [00:07:46].
Future Focus Areas
Looking forward, Alma is doubling down on three key areas [00:07:54]:
- Brand and Design: Code is becoming a commodity, and a user-centric product with visually appealing design is crucial for standing out [00:07:59].
- Trust and Partnerships: Earning user trust, especially as a new entity, involves partnering with organizations and individuals that users already trust and who share an aligned mission [00:08:26]. Alma actively asks users who they trust for information and seeks partnerships [00:08:36].
- Community and New Data: As the community grows, there’s interest in how others eat and learning from their habits [00:08:52]. Alma is focusing on features that leverage the community to create net new, valuable data, recognizing that major LLM companies are already “hoovering up” online data [00:09:16].
The development of the Alma score for food quality assessment, a metric out of 100 developed with academic advisor Dr. Eric Prim, also emerged from user feedback indicating a desire to understand holistic eating quality beyond just calories and macros [00:02:14].