High quality, cost-effective genAI applications are challenging to build. One big reason is that systematic methods for tuning, testing, and optimizing are still being refined.
GenseeAI is a unified AI workflow problem that solves that problem – it enhances AI output quality and reduces costs at the same time.
GenseeAI uses the Cognify optimizer – an open source tool released by UC San Diego AI researchers Yiying Zhang, Reyna Abhyankar, and Zijian He.
Cognify results on genAI performance:
✅ Improves gen-AI applications’ quality by up to 48%
✅ Reduces gen-AI applications’ execution costs by up to 9 times
✅ Works seamlessly with Langchain, DSPy, and Python programs
We got an early look at GenseeAI and Cognify with Dr. Yiying Zhang and Reyna Abhyanker before they released Cognify to the world in late November 2024.
Here’s a live presentation from the Codingscape AI practice to show us how their workflow optimizer gets such significant results and what to expect from the full platform.
Watch GenseeAI live talk
Solve major challenges with building AI workflows
If you're working with AI in production, you're likely familiar with the challenges. Most deployments aren't just single models but complex workflows involving multiple LLMs, data sources, tools, and sometimes human input.
Current solutions require manual workflow construction, tedious tuning, and troubleshooting when things go wrong. Not to mention the sky-high costs – often 10x to 100x higher than necessary.
Enter Cognify workflow optimizer
Think of Cognify as your AI efficiency expert. It takes your existing workflows (whether written in Langchain, DSPy, Python, or other frameworks) and applies a two-stage optimization process.
- Graph structure optimization – First, it analyzes your workflow structure, identifying critical steps and reorganizing for efficiency. It can even parallelize independent operations automatically.
- Model selection – Then, it selects the most appropriate models for different steps, balancing cost against quality using sophisticated Bayesian optimization.
The results are impressive: up to 48% quality improvement and 6x-9x cost savings. Even better? The optimization process itself is surprisingly affordable, typically costing just 50-80 cents.
You can find Cognify on Github to start improving your generative AI workflows.
Next up: GenseeAI platform with Serving System
Cognify is a major part of the GenseeAI platform and the Serving System is in development to be released soon. The serving system makes it easy to run genAI workflows in real-world, multi-tenant environments.
It includes:
- Distributed workflow scheduling
- Auto-scaling capabilities
- Intelligent load balancing
- Sophisticated state management
- Resource isolation
By automating optimization and improving serving efficiency, GenseeAI addresses many of the pain points organizations face when deploying AI solutions.
While it's still in development, the early results are promising. GenseeAI represents a significant step forward in AI infrastructure. It has the core functionality to become an essential tool for organizations looking to make their AI deployments more efficient and cost-effective.
Start working with Cognify, the workflow optimizer and stay tuned for the release of the full platform. For organizations heavily invested in AI workflows, GenseeAI is a platform worth watching.
