TikTok Pulse

• 

  home page

Inspiration

With the massive growth of TikTok Shop, we noticed a gap in the fitness and wellness category: users often watch fitness content or health advice but struggle to bridge the gap between their personal nutrition tracking and purchasing the right products to hit their goals. We wanted to build an intelligent bridge between personal health tracking and social commerce.

What it does

TikTok Pulse is a multimodal AI sports nutrition coach integrated directly with the e-commerce ecosystem. Users can simply take a photo of their meal and drag it into the app. Using advanced Vision AI, TikTok Pulse scans the food, instantly estimates the macros and calories, logs them against the user's daily budget, and acts as a personalized coach.

But here is where the e-commerce magic happens: when the AI notices a nutritional deficiency (e.g., "You are 40g short on protein today"), it seamlessly recommends highly-rated wellness products, supplements, or gear available directly on TikTok Shop.

How we built it

Frontend: We built a premium, glassmorphism-styled client-side application using HTML5, Vanilla CSS3, and JavaScript to match TikTok's modern, engaging aesthetic. Multimodal AI: We integrated the Llama 3.2 Vision Instruct model (via Fireworks AI) for blazing-fast visual parsing of food photos. Conversational Coach: We used Llama 3.1 70B as the underlying coaching engine, feeding it the user's live biometric context (BMI, age, weight goals, daily macro logs) so it can provide hyper-personalized advice and product recommendations. Architecture: The app uses secure localStorage to cache user profiles and API keys, ensuring zero latency, offline capabilities for data charts, and complete privacy.

Challenges we ran into

Getting the Vision model to accurately estimate portion sizes and caloric density from a single 2D image was difficult. We overcame this by writing a robust meta-prompt that forces the AI to break down the plate into visible ingredients, estimate their volume in grams, and calculate macros systematically before returning a structured JSON response to the frontend.

Accomplishments that we're proud of

We are incredibly proud of the UI/UX. We managed to build a gorgeous, interactive HTML5 canvas for the 7-day trend analytics without relying on heavy charting libraries. Additionally, having a true multimodal drag-and-drop interface running flawlessly entirely on the client-side feels like magic.

What we learned

We learned a massive amount about prompt engineering for Vision models. We also gained deep insights into how context windows can be effectively managed on the client-side by maintaining a rolling window of the user's most recent meals rather than sending their entire lifetime history to the LLM.

What's next for TikTok Pulse

TikTok Shop API Integration: Turning the AI's product recommendations into one-click checkout buttons using the official TikTok Shop APIs. TikTok Video Export: Allowing users to generate a customized TikTok video of their weekly fitness progress chart to share on their feed natively!

Built With

• artificial-intelligence
• css3
• e-commerce
• html5
• javascript
• llama3
• machine-learning