THE AI ECOSYSTEM
Bridging RAG (Retrieval-Augmented Generation) with User Centricity.
Overview of Foodhak Assistant / AI Chatbot • Frameworks • Agents Orchestration • Logic: Insights (Reward)
The Problem
- Users felt overwhelmed by generic health advice
- Overwhelming food choices / unaware of nutrients needed
- Exhaustion on downloading too much apps
- Another Chatbot feature

Chatbot experimental feature 1.0
- User query also includes User input on Images and pdf inputs. (blood works, pathology reports)
The Idea
We aim to designed a "lifestyle companion" AI product. Instead of a blank chat box, we’ve implemented Contextual Triggers the AI proactively surfaces insights based on wearable data (e.g., "Your glucose peaked after lunch; here's why…"). (”You should reduce your sodium intake since you’ve been munching on XYZ past 3 days”). (”If you walk an extra 10 mins today, we could reduce X weight on weight day…”).
- We needed to transform static health data into actionable, safe, and personalized dialogue.
- Be insightful, hyper-personalised health advice or actions - find ways to streamline to their lifestyle, enable delights and optimism.
High Level Technical Deep-Dive
- Our AI chatbot was built on a RAG architecture. This ensures that the provided responses cited is clinically-backed and personalised to user. Thus, eliminating the risk of inaccurate responses and recommendations, building user’s trust.
- Using 3rd party *Terra API for activities tracking, syncing major apps like (fitbit, strava, apple, google tracking)
- *Hyper-localisation and real-time RAG for precise information and immersive experience.
The Impact
- +40% increase in weekly active users (WAU) interacting with the Assistant. especially on meals logging feature.
- User also recur due to its anticipating daily insight. (e.g., sleeps and meal intake analytics of yesterday and week’s summary)
Setbacks
- generation took time, a-sync was introduced, due to tech limitations. was planning to reduce time to insight from 3 mins to <10 seconds.