Introducing AI Builder: the A.I engine for building intelligent document applications
May 25th, 2023
As AI models are becoming commoditized, high quality training data is becoming key to successful and applicable AI. chatGPT is the perfect example, feeding GPT-3 with a small amount of high quality human labeled dataset using RLHF, OpenAI created on of the most advanced AI model available to humanity. At UBIAI, we have known the value of high quality data since day 1. Since the launch of our data labeling and model training platform in 2021, we have received tremendous traction across multiple industries, leveraging our platform to create high quality training data that powers high performance NLP models.
However, over the past year, we have witnessed two re-occurring issues across our customers:
How do I deploy my fine-tuned model without investing in a large MLOps team?
How do I create custom workflows with my trained model that works for my business without spending hundreds of thousands of dollars and months of building an application.
This got us thinking, what if you can deploy fine-tuned models with just few clicks but also integrate them into a custom workflow that is tailored to the company’s needs? And this is where the idea of the AI builder came from. We are excited to announce today the release of AI Builder in Beta! Here is a sneak peek:
The Need for Customized Solutions
Businesses that deal with a large amount of unstructured data often face significant challenges in processing and extracting valuable insights. Off-the-shelf intelligent document processing (IDP) solutions offer static workflows with very limited capabilities in terms of model and workflow customization.
For example, most IDP solutions will provide document upload -> OCR -> Entity Extraction -> Document classification -> Export, but what if you need to add relation extraction, summarization, zero-shot classification, or question answering to your workflow? Tough luck, you either need to create your own application from scratch or pay for a hefty sum for a custom feature development.
This is where AI Builder comes in. To address this issue, we have built a versatile application that is modular in nature that allows you to create custom applications.
AI Builder custom workflow creation
It is essentially comprised of data processing modules that you can logically connect sequentially or in parallel to create what is called a workflow. Users can combine different modules together to create a workflow that is tailored to their business need. The use cases range from intelligent document extraction from bank statements, invoice, receipts to summarization and synthetic data generation.
AI Builder has been built on three principles:
- Easy deployment of fine-tuned models with just few clicks
- No-code custom workflow creation
- Intuitive and easy-to-use UI
Custom Workflow Creation
No more getting stuck with off-the-shelf one-size-fits all workflows. With the AI Builder, you can build your own workflow easily by logically connecting different NLP modules using intuitive drag-and-drop interface. The modules act as the building blocks of your custom workflow.
The modules are categorized by functions: Import, Processing and Export.
1. The import modules allows you to import all type of unstructured data such as PDF, JPG, PNG, DOCX, TXT, HTML and CSV files
Data import modules
2. The processing modules allows to customize your processing pipeline using a variety of services:
- OCR: Parse text from PDF and scanned images
- Custom Form Recognition model: Plug in your custom form recognizer model, suitable for semi-structured documents such as invoices, receipts, tables, contracts with layout dependency
- Custom NER model: Plug in your custom NER model, could be fine-tuned spaCy, Bert or LayoutLM model
- Custom relation extraction model: Plug in your custom relation extraction model based on spaCy
- Table extraction module: Extract any table from your PDF and images
- chatGPT module: Prompt chatGPT with instruction to act on your data
- Pre-built models: invoice, ID and receipt
Data processing modules
We are adding more modules every week, so if you have any new module you want to add, please shoot us an email!
3. Export: once the data is processed you can select the type of export API, JSON or Excel.
Data extraction in AI Builder
One-click Deployment of Your NLP Model
No more hassle deploying your NLP Model! With AI Builder one click deployment, simply port your trained model from UBIAI Platform or upload your own model directly and get an API endpoint right away.
AI Builder allows businesses to create custom NLP models that are specifically tailored to their unique requirements. This means that businesses can deploy highly accurate and effective models that are optimized for their specific use case with few clicks.
Custom NLP model selection
Integration with LLM
LLMs are becoming ubiquitous in everyday business operations. With AI Builder, you can easily integrate LLMs like chatGPT within a custom workflow. Simply drag and drop the chatGPT module and connect your input data. By connecting LLM to user’s own custom model and data, we can achieve significant improvements in various use cases such as entity-based text generation, synthetic data generation, summarization, custom NER + zero-shot document classification, model prediction verification, etc.
LLM prompt windows
Pre-built NLP Templates
AI Builder provides a library of pre-built document processing templates so you can start getting actionable insights from your unstructured documents immediately.
AI Builder pre-built templates
Human-in-the-Loop
AI Builder also features human-in-the-loop, which allows users to review, correct model predictions and re-train the model to improve accuracy and efficiency. By incorporating human feedback, the fine-tuned model can learn and adapt to the specific needs of each business, ensuring that the resulting NLP workflows are highly accurate.
AI Builder is specifically designed to help businesses develop customized NLP workflows that are tailored to their specific needs. This is achieved through a combination of machine learning algorithms and human-in-the-loop.
Human review of semi-structured documents
Human review of unstructured documents
Use Cases
The use cases that can be achieved are numerous, here are few of them:
- Custom intelligent document extraction from bank statements
- Custom intelligent document extraction from insurance documents
- Intelligent document extraction and classification of receipts and invoices
- Logistics and transportation billing extraction
- SEO optimized product description generation
- Entity and relationship extraction from scientific articles
- Entity based content generation
- Synthetic data generation
- Entity-based Sentiment analysis from financial news articles
Conclusion
Over the past 6 months, we have been working tirelessly with our customers to develop AI Builder, a cutting-edge solution that has the potential to revolutionize intelligent document extraction from unstructured data. With this launch, businesses can now easily deploy their custom NLP models and integrate them seamlessly into a data processing workflow without any code required.
We invite you to experience the capabilities of AI Builder by scheduling a personalized demo.