Popstock: AI Financial Literacy Through Pop Culture
Success Stories
Industry: Edtech
About customer:
Popstock Educational Services teaches financial literacy and data analysis through a simulated stock market blending pop culture. Using AI-driven tutoring and videos, it simplifies stats and probability, focusing on underprivileged youth to equip them with vital skills for financial success.
The challenge:
Popstock aimed to make finance fun by linking it to pop culture, but needed an AI to answer student queries like “Should I invest in Taylor Swift?” They faced challenges in training LLaMA 2 with custom data, ensuring secure, scalable Google Cloud integration, and building an engaging app while keeping costs low and data private for reliable, real-time classroom use.
The solution:
Evonence fine-tuned LLaMA 2 on Google Cloud’s T4 GPU using LoRA, processing custom pop culture and finance data stored in Google Cloud Storage. Also built a Streamlit app for students to ask questions, ensuring a secure, automated, and scalable system that delivered fast, relevant answers tailored to classroom learning.
Leveraging Google’s product suite:
Google Cloud Storage, T4 GPU, Vertex AI, Compute Engine
Cloud scale and speed:
The solution is designed for cloud scale and speed by deploying the fine-tuned LLaMA-2-7B model on a Vertex AI Endpoint. This managed service is configured for auto-scaling, allowing it to automatically adjust the number of serving instances based on the volume of student queries. This ensures a responsive experience during peak classroom hours while conserving resources during off-peak times. The training process was also optimized for speed and efficiency, using NVIDIA Tesla T4 GPUs and techniques like mixed-precision training (FP16) to significantly accelerate the fine-tuning of the large language model.
Google AI-enhanced predictions:
The system provides Google AI-enhanced predictions through a LLaMA-2-7B model that has been specifically fine-tuned to understand and generate responses that connect financial concepts with popular music artists. The model was trained on a custom Q&A dataset stored in Google Cloud Storage, teaching it PopStock's unique domain knowledge. When a student asks a question like, "Should I invest in Taylor Swift's stock?", the system uses the fine-tuned model to generate an immediate, relevant, and easy-to-understand response. The model's grounding in this custom knowledge base allows it to explain complex financial ideas using engaging and relatable pop culture examples.
Partner role in the project:
Evonence has built an AI system for Popstock, fine-tuning LLaMA 2 on Google Cloud with a T4 GPU. Also managed data storage, created a user-friendly web app for student queries, and ensured a secure, scalable setup for engaging, tailored educational insights.
The results:
The fine-tuned LLaMA 2 model powered a web app that gave students instant, engaging answers like “Beyoncé’s stock is trending up!” It made finance fun, saved teachers time, and scaled securely on Google Cloud, enhancing classroom learning with reliable, pop culture-infused insights ready for future growth.
High availability:
High availability and resilience are ensured by deploying the fine-tuned model to a Vertex AI Endpoint, a managed service designed for robust, continuous operation. Vertex AI automatically handles infrastructure management, health checks, and replication. The endpoint is configured with a minimum number of replicas to handle requests even if a single instance fails, and it can auto-scale to meet fluctuating traffic demands. This managed, scalable infrastructure ensures the PopStock smart assistant remains available and responsive for classroom use.
AI-based predictions:
The AI-based prediction system functions as a "smart assistant" that provides instant, understandable answers to student questions about investing in music artists. After being fine-tuned on PopStock's custom data, the model can generate new, relevant responses on the fly. For example, students can ask questions like, "What's been happening with Beyoncé's stock lately?" or "If Drake releases a new album, how might that change his stock price?" and receive an immediate, AI-generated response that explains the financial concept in a way that resonates with their interests. The goal is to provide educational explanations, not direct financial advice.
Related Success Stories
