Summary
As part of the GCP GenAI Sprint Program, our cross-functional Presidio teams collaborated with Google Cloud experts to rapidly design and prototype generative AI solutions that solve real-world business challenges. Over a few focused weeks, we built and tested production-ready use cases using Google Cloud’s Vertex AI platform—delivering immediate, measurable impact.
Here’s a look at three powerful examples:
1. Sentiment Analysis at Scale
Understanding customer sentiment is vital—but doing it at scale requires more than keywords. We built a web application using Python and the Streamlit framework to explore three approaches to sentiment analysis:
- Natural Language API: Fast and scalable with minimal setup—great for general insights, though less effective for niche domains.
- Vertex AI Custom Model: Tailored to your specific language and industry. More accurate and transparent, but requires time to train.
- Gemini for Google Cloud: The most advanced option—multilingual, context-aware, and ideal for nuanced, real-time sentiment detection.
Result: Gemini outperformed across scenarios including sarcasm, informal tone, and multilingual content—proving ideal for customer service, reviews, and social listening.
2. Sales Forecasting with ARIMA and LSTM
Forecasting sales accurately is key to smarter inventory planning and marketing. We built an end-to-end solution that:
- Ingests raw data such as transaction date, item category, product name, and item price from historical sales records in cloud storage.
- Trains and tunes ARIMA and LSTM models using Vertex AI.
- Automatically deploys the best-performing model for real-time or batch predictions.
- Enables continuous retraining when new data arrives.
Result: Fully automated, scalable forecasts delivered through a clean front-end—empowering sales teams to respond to trends in near real time.
3. Custom Image Classification with Vertex AI
Image classification unlocks huge value, whether you’re detecting product defects, identifying plant species, or anything in between. The goal was to classify images into multiple categories. Starting with this clearly defined objective and a structured
dataset, we could build a focused and relevant model.
The dataset included labeled images in distinct classes, eight unique categories of classification, and images in compatible JPEG/PNG formats.
We tested:
- AutoML: A quick, no-code path that delivered 90% accuracy with minimal effort.
- Custom EfficientNet Model: Gave us full control and flexibility for complex multi-label tasks.
Result: AutoML offered speed and performance out of the box. Custom models provided more flexibility and transparency—ideal when precision and customization are key.
Ready to Innovate with GenAI?
These use cases are just the beginning.
At Presidio, we are constantly innovating to make real GenAI-driven impact for our clients. Whether you’re looking to better understand customers, predict trends, or extract value from images, our GenAI experts can help you build, deploy, and scale intelligent solutions, fast.
Let’s talk about what’s possible for your business. Contact us today.
Manojkumar TS
