Ethan Lab Software Development · Product Management · Tech Reviews
Back to research notes

[Conversational AI-8] Introduction to R&D Departments and Development Processes

Updated

This article uses the company where the author works as an example; without disclosing company confidential information, it introduces the ‘relevant departments, development processes, and job responsibilities’ involved in AI product R&D, unveiling the mystery behind the AI industry.

The company’s main business involves introducing AI technology, Natural Language Processing (NLP) technology, system integration, and platform construction for sectors such as telecommunications, finance, government, and e-commerce. It most commonly helps large institutions build systems like Chatbots and call centers, assisting in improving customer service and reducing labor requirements.

Organizational Structure of an AI Company

The R&D center of a typical AI company can be divided into the following departments (as an example). The horizontal axis represents department names, and the vertical axis represents the organizational structure:

  • Research Institute / Knowledge Research Department (Theoretical and Algorithmic Research)
  • Natural Language Understanding (NLU) R&D Department / Speech Engine R&D Department (Algorithm Research and Application)
  • Core Product R&D Department / Cloud Platform R&D Department (Engine Application)
  • Product Testing Department
  • Solutions Department (Product Packaging)
  • Project Implementation Department (Product Application and Secondary Development)

Development Process for AI Products

  1. The Research Institute is responsible for collaborating with universities, reading papers, and researching cutting-edge technologies, then providing the research results to the NLU / Speech R&D Department. At the same time, it notifies the patent application unit to register valuable technologies as patents.
  2. The NLU / Speech R&D Department turns research results into underlying engines and hands them over to the Core Product R&D Department. At the same time, it notifies the patent application unit to register valuable technologies as patents.
  3. The Core Product R&D Department is responsible for market research, product design, and writing front-end and back-end logic, packaging the engines into products such as Chatbots and Recommender systems. At the same time, it notifies the patent application unit to register valuable technologies as patents.
  4. The basic content and industry knowledge for Chatbot / Recommender products are built by the Knowledge Research Department, such as data annotation, knowledge base construction, knowledge graphs, and pre-set industry packages.
  5. The Solutions Department is responsible for packaging the various capabilities of Chatbot / Recommender products into industry-specific solutions, providing them to the sales unit for selling.
  6. If a customer is interested, the Project Implementation Department uses the Chatbot / Recommender products / solutions to conduct a free POC (Proof of Concept) for the customer; if the customer is satisfied with the results, they will sign a contract and pay for subsequent development.
  7. The Project Implementation Department then completes the system construction and secondary development work, such as API integration for Chatbot / Recommender systems, based on the contract terms, and provides after-sales maintenance services. The company then charges annual License, maintenance, and re-development fees.

Job Responsibilities in the AI Industry

If you are interested in entering the AI-related industry, you can pursue the following departments, but the technologies emphasized differ:

Research Institute

Generally uses the Python language and frameworks such as TensorFlow, PyTorch, and SKlearn to research algorithms, such as Automatic Speech Recognition, Word Segmentation, Part of Speech tagging, Syntactic Analysis, Semantic Analysis, Dialogue Management, Natural Language Generation, and Text to Speech.

NLU / Speech R&D Department

Since Python is not suitable for large-scale parallel processing, Java is commonly used here, along with frameworks like Mahout and Deeplearning4j, to implement algorithms, engines, training, and model deployment on Hadoop and Spark.

Core Product R&D Department

Here there are fewer limitations; any language and framework can be used. For example, using the Go language and the Gin framework to write back-end logic and package underlying engines into products; and using the JavaScript language with frameworks like React or Vue to develop the front-end.

Project Implementation Department

Selects languages and frameworks based on customer needs, conducts secondary development for customers using Chatbot / Recommender products / solutions, and provides after-sales maintenance services.