10. TML Solution Components

Below describes the entire TML technology solution stack.

10.1. 1. TML Components: Three Binaries

TML is comprised of 3 binaries written in Go: found on Github

  1. Viper - source - sink binary for Apache Kafka - runs on MAC/Windows/Linux

  2. HPDE - AutoML binary for real-time data - runs on MAC/Windows/Linux

  3. Viperviz - Visualization binary for real-time dashboards - runs on MAC/Windows/Linux

Important

These 3 binaries perform all functions in every TML solution. These binaries can be seen as microservices that can be instantiated any number of times to scale your solution for unlimited data processing.

These binaries are critical for TML solutions and are the “secret sauce” inside TML.

Binary

Description

Viper

This is the CORE binary that performs

all the major TML functions of processing

and machine learning. This binary acts like a

microservice that can be

instantiated any number of times to process

large amounts of real-time data.

This binary is compatible with REST API. TML

solutions (using Python) connect to this binary

and instruct it to stream data to Kafka, preprocess

data, and

develop training data sets for machine learning.

HPDE

Hyper-prediction technology performs all

machine learning functions. Viper connects

to HPDE, using REST, and instructs it to

perform machine learning. By

off-loading this function to HPDE, TML can

perform very fast machine learning (in few seconds)

for each entity and sliding time window.

Refer to :ref:`TML

Performs Entity Level Machine Learning and

Processing` and Entity Based Machine Learning By TML

Viperviz

This binary performs real-time streaming

visualization using websockets. This is a very

powerful binary because it uses the underlying

network to streaming

data to a client browser for fast, and very

cost-effective, visualization of real-time

TML solution outputs. This means users do

NOT need a third-party

visualization tool like Tableau or PowerBI. Users

can create amazing real-time dashboards

quickly and cheaply. Refer to example dashboards

here TML Real-Time Dashboards

10.2. 2. TML Component: One Core Python Library

MAADSTML Python Library: https://pypi.org/project/maadstml/

MAADSTML Python library : API to build TML solutions that connect to the Viper binary. Refer to the MAADSTML API here MAADSTML Python Library API

Important

All TML solutions use the the MAADSTML python library. This is a critical library, and controls the 3 binaries.

10.3. 3. TML Component: Apache Kafka

TML integrates with Apache Kafka - on-premise or in the cloud.

Attention

TML binaries are integrated with Apache Kafka. TML solutions can be run On-PREMISE using Open Source Kafka or in the CLOUD using AWS MSK or Confluent Cloud.

10.4. 4. TML Component: Docker Containers

All TML solutions are containerized with docker for production deployments.

10.5. 5. TML Component: Kubernetes

All TML solution containers scale with Kubernetes. This allows companies to build fast, scalable, real-time solutions.

10.6. 6. TML Component: PrivateGPT for Generate AI

TML uses PrivateGPT for fast, real-time, AI. The container is here Docker PrivateGPT

Refer to TML and Generative AI for more details.

10.7. 7. TML Component: TML Solution Studio Container

For convenience, TML solution can be easily built using the TML Solution Studio container. Refer to TML Solution Studio (TSS) Container for further details.

10.8. How The TML Components Are Integrated

TML solutions are developed using the MAADSTML Python library that connects to the TML Binaries, using REST API, for streaming real-time data to Apache Kafka, processing data in Kafka, and performing machine learning. Once the TML solutions are built, they are containerized with Docker and scaled with Kubernetes.

Attention

TML performs in-memory processing of real-time data and does NOT require an external database - ONLY KAFKA is needed. This results in dramatic cost- savings for storage, compute and network data transfers.

TML does NOT perform SQL queries, it performs JSON PROCESSING. This results in much faster, and much cheaper processing of real-time data.