Use Cases implemented using Trillo Workbench

Enterprise data resides in multiple systems. Trillo Workbench can ingest data from these systems into BigQuery, Cloud SQL, Cloud Storage Bucket, and Snowflake. A complete enterprise modernization entails the following functions.

1. Use third-party services, such as Fivetran, Stitch Data, etc., to pull data from the source system into BigQuery or Cloud SQL.
2. Write custom connectors as Trillo Workbench workflows when the data ingestion requires complex processing, such as transforming data before loading, event-driven decisions, etc.
3. Transform data once data is available in the BigQuery tables or Cloud Storage Bucket.
4. Provide master data ingestion UI from Excel/CSV/JSON/XML files or manually.
5. Role Based Access Control (RBAC) rules for governance and compliance.
6. Data catalogs using Trillo Workbench and Google Dataplex.
7. Data security measures such as column encryption and redaction.

A data warehouse for healthcare requires similar functions as an enterprise data warehouse described above. It has the following differences.

1. It requires Trillo Workbench connectors due to their non-availability, and if FHIR/HL7v2 compliance is required.
2. Data warehouse should be HIPAA compliant.

Trillo helped several Google customers modernize their on-premise applications and migrate to the cloud. The applications covered under this use case had the following characteristics. This is an ongoing use case we still fulfill for new customers.

1. They were legacy applications running on-premise or colo centers servers.
2. The applications used old Microsoft technologies or a very old Lamp stack.
3. The customer managed the data server (backup, restore).
4. These applications were at the end of life and performing poorly. A few customers reported one failure per month.

Trillo modernized these applications using Trillo Workbench in a few months and at a low cost. Anything else would have cost the customers a year or more work. It involved the following steps.

1. Understand existing applications and propose new designs through a reusable POC (using Trillo Workbench as the server).
2. Design a data model using Trillo Workbench UI.
3. Write application logic as serverless functions.
4. Replaces existing databases with the managed Cloud SQL instances. Write a workflow for data migration.

The telehealth application enables collaboration between patients, providers (doctors. nurses), pharmacies, labs, and facility administrators. For example, a patient can search a provider, set up appointments, browse visits and lab reports, etc. A provider can note visits, send prescriptions to a patient’s pharmacy, and send lab orders. Providers and patients can virtually meet by phone or video conferencing. The system runs background tasks for transcribing audio files recorded by providers, downloading reports from labs, generating internal reports, transferring data to a data warehouse, etc.

For implementation, the telehealth application is a special case of the use case discussed above. It has the following specificity.

1. The data model is specific to Telehealth and has entity types such as Patient, Encounter, Appointment, Lab, etc.
2. The workflows are specific to healthcare, such as transcribing, downloading reports, etc.
3. Serverless functions are specific to the domain, such as Save Encounter, Send Prescription, etc.

This use case is a part of the solution Trillo built for a brokerage firm. The brokerage firm sells products from manufacturers to grocery shops. Each manufacturer allocates a certain quota of products to it. Its sales team sells them by phone, email, or in person. During the interaction with a customer, a salesperson needs to answer certain questions, such as,

1. How many boxes of “diet coke” are in stock?
2. What is the price of a box of “diet coke”?
3. Is there a promotion available on “Florida orange juice”?

The above use case is served by an application providing a conversational interface on mobile devices and computers. The application flow at a high level is as follows:

1. Using a web or mobile application, a user asks a question.
2. The question is converted to text using Google Cloud’s speech-to-text service.
3. The text is passed to an AI service to detect intent and entities within questions such as “product = diet coke”, “query = stock”, etc.
4. The returned result of intent is processed by application logic which substitutes template values in the answer with actual values queried from the database.
5. The final text of the answer is converted to an audio file using Google Cloud’s text-to-speech service.

This application uses semantic matching to match jobs with resumes or a resume with other resumes. Semantic matching provides high-quality matches ranked by score compared to keyword-based searches. The application has two main processes, as follows.

1. Indexing Resumes and Jobs: It involves following steps.
   1. Extract text from resumes and jobs provided as pdf or doc files.
   2. Parse text to categorize content such as education, experience, locations, years in job, companies, schools, etc.
   3. Generate vector representation of text using fine-tuned NLP models.
   4. Store vectors in Vertex AI Matching Engine. We store job and resume vectors in separate vector databases.
2. Semantic Matching: Matching input is a document. It involves following steps. The first three steps are similar to indexing.
   1. Extract text from resumes and jobs provided as pdf or doc files.
   2. Parse text to categorize the content.
   3. Generate vector representation of text.
   4. Match the input document’s vector against the vector database for similarity.
   5. Sort results based on the matching score and return the result.

In 2018 and 2019, we helped a customer with a very large-scale Kubernetes (GKE) deployment. The purpose of this application is to automate telecom mobile service sales processing. It required validating the purchaser’s identity in near real-time by processing a photo and documents provided by the purchaser. The process is automated using Vision AI.

This application is built using open-source libraries. It does not use Trillo Workbench. It used a set of microservices for business process orchestration, data archiving, and AI model serving. The main challenge was designing high-performance microservices architecture and utilization of GPUs for Vision AI models. It took about a year to optimize the application for high performance (not counting the time to build AI/ML models). 

Looking back, we can deliver a similar application using Trillo Workbench in 2-3 months. It will perform better, be more robust, secure, and self-healing.

Trillo File Manager enables this use case and delivers the following functionality.

1. SFTP.
2. Web UI for managing files and folders securely.
3. Share files with customers, partners, and team members.
4. Group folders for collaboration.
5. Audit trail for compliance.
6. HIPAA compliance.
7. Extensible to support business process automation.
8. Deployed in your GCP environment ensuring 100% data protection.
9. Scalable using GKE version.