From: aidotengineer
Alma’s core belief is that eating well should not be difficult, and good health stems from simple, personalized nutrition [00:00:10]. Traditional nutrition apps often demand extensive information from users while providing minimal value in return [00:00:34]. Alma aims to solve this with AI, envisioning a true nutrition companion [00:00:44].
The vision for Alma’s AI nutrition companion rests on three pillars:
- Simple Nutrition Tracking: Making tracking easy, natural, and conversational, like texting a friend, rather than tedious searches or unreliable photo analysis [00:01:04].
- Contextualized Insights: Building strong, powerful context around the user’s preferences, habits, and interests to guide them towards their goals [00:01:23].
- Product Connection: Connecting users with appropriate products, restaurants, and meals to help them achieve their health objectives [00:01:37].
Alma also developed the “Alma Score,” a score out of 100, in collaboration with Dr. Eric Prim at Harvard University [00:02:31]. This score helps users understand the holistic quality of their diet, nudging them towards beneficial foods and away from less healthy options [00:02:38]. This concept emerged from early user feedback indicating a desire to understand food quality beyond just calories and macros [00:02:11].
The Pivotal Role of User Feedback in AI Development
A significant lesson learned during Alma’s beta phase (a closed beta for about 4 months, shipped in February) was the critical importance of relying on user feedback [00:01:08], [00:02:08], [00:03:01]. While AI evaluation methods are explored, nothing surpasses real-time, in-the-moment feedback from users [00:03:06].
Directly Soliciting Feedback
One of Alma’s key best practices for gathering feedback is a pop-up toast notification that asks, “How did Alma do?” after every interaction [00:03:17]. This feature has been invaluable in guiding development and measuring accuracy and improvements [00:03:29].
Lessons Learned from User Interactions
Through consistent user engagement, Alma has gained several insights influencing its development process:
- Constraining LLMs: Large Language Models (LLMs) tend to have high error rates when given open-ended, very large tasks [00:03:39]. It’s crucial to constrain the LLM to specific tasks [00:03:48]. Alma addresses this by breaking down the information relay process into steps, sending down certain aspects at each stage (e.g., recognizing a food item, extracting it, sending it to the client, then processing it with a USDA database) [00:04:08]. This approach makes the experience feel much faster and more interactive for users [00:04:46].
- User Value of “Streaks”: A bug that broke user “streaks” (consistent tracking) caused significant outcry in Alma’s WhatsApp community [00:05:03]. This revealed that a feature initially considered minor was highly valued by users, prompting Alma to double down on it [00:05:21].
- Building User Context: Users become frustrated when they have to reiterate information [00:05:41]. Alma addresses this by constantly learning about the user; any novel piece of information from an interaction is added to a knowledge dataset about the user [00:05:58]. Users can view, delete, or add to this information, ensuring Alma gets smarter with every interaction [00:06:06].
- Proactive Insights: Users don’t always open the app to ask questions [00:06:16]. Alma therefore proactively detects insights about users’ eating habits and surfaces information they might not know [00:06:30]. This creates a constant, bite-sized learning process for the user [00:06:54].
- Multimodality: While some developers might prefer a specific interaction modality (e.g., voice), user feedback showed that users appreciate the flexibility of multimodality [00:07:01], [00:07:27]. Alma makes it easy to track food via voice, photo, or text, allowing users to choose based on context [00:07:35].
Looking Forward: User-Centric Strategic Pillars
Alma’s future strategy is heavily influenced by these user-centric learnings:
- Brand and Design: As code becomes a commodity, strong brand identity and visually appealing, user-centric design are crucial for standing out [00:07:59]. This is ingrained in Alma’s DNA [00:08:14].
- Trust and Partnerships: Earning user trust, especially as a new product, involves partnering with entities users already trust and aligning on mission [00:08:26]. Alma actively asks users about their trusted information sources to explore potential collaborations [00:08:36].
- Community and New Data: As the community grows, there’s user interest in how others eat and a desire to learn from shared habits [00:08:52]. Since major LLM companies are “hoovering up” existing online data, Alma focuses on leveraging its users to create valuable, net new data [00:09:23].