Ml: a detailed introduction to the division of the top ten roles, backgrounds, responsibilities and outputs of the machine learning engineering team

ML： Background of top ten roles of machine learning engineering team 、 duty 、 Detailed introduction to the division of output

background ： In the process of building a machine learning team , We need systematic engineering thinking and role setting , To avoid as much as possible “ Technical debt ” And “ Lack of soft power ” Cause potential “ Technological disaster ” And “ Deliver disaster ”.

Catalog

Background of the top ten roles of machine learning team 、 duty 、 Division of output

1、Product Manager/ The product manager —A1 role

2、Project Manager/ project manager —B1 role

How to quickly realize commercialized products ？

3、Business Consultant/ Business consultants ( coordination PM)— Role concept —A2 role

4. Data Scientist/ Data scientist —C1 role

5. ML Researcher/ Machine learning researcher —C2 role

6. Data Engineer/ Data Engineer —C3 role

7. ML Engineer/ Machine learning Engineer —C4 role

8. DevOps Engineer/DevOps The engineer —C5 role

9、Software Develop Engineer/ Software Development Engineer —D1 role

10. Delivery Engineer/ Delivery Engineer —D2 role

---

Background of the top ten roles of machine learning team 、 duty 、 Division of output

Minimum Viable Product,MVP, The simplest practical product .

1、Product Manager/ The product manager —A1 role

background	The machine learning algorithm consists of a “ Technical achievements ”→“ Which can be delivered on the ground “ product ”, While meeting the functional requirements of customers , Within the range of acceptable indicators （ precision 、 Speed etc. ） The commercial goal of stable delivery and operation within the company .
duty	(1)、 Keep up with industry trends , Combined with the company's capabilities , Explore product innovation (2)、 Provide product solutions ： Responsible for specific product design , Research and analysis of competitive products , Put forward optimization scheme and design product function based on operation data , Offering products / Project solutions (3)、 Preach and exchange with customers 、 Preparation of bidding documents ： Cooperate with sales / Front line roles such as solution architects , (4)、 product / Project implementation and delivery ： Achieve business income (5)、 Product iteration management , Enhance user experience
Main work outputs	(1)、PRD Product requirements document ： Including flow chart, etc (2)、 Product white paper / Pre sales documentation 、 Bidding documents (3)、 Project solutions
Excellent qualities	(1)、 insight ： There is no more scientific “ Prior knowledge ”, Most of them can only be tested through the product results . Improve insight , There is no shortcut. , Only a deep understanding of the target area , Keep thinking 、 Think deeply , Rehearse over and over again . Demand for mining ： How to discover “ potential ” demand , A lot of times “ Implicit demand ” Or more valuable . Demand for judgment ： distinguish “ Pseudo demand ”, Look for the “ The essence ”. Demand analysis ： Behind the demand “ motivation ” What is it? , The needs of the “ scene ” What is it? , The needs of the “ Cause and effect chain ” What is it? . User identification ： Is it right “ real ” user ,“ Typical users ” Whether the requirement has “ Pervasive ” sex . value analysis ： Requirements are realized to users 、 The value of the product , The value to users after the product is realized . (2)、 Risk predictability ： What are the consequences of model mispredictions ？ risk analysis ： The cost of requirements realization 、 Cycle and demand fade cycle . (3)、 Sensitivity ： Sensitive to the user's product experience
Workflow	(1)、 Demand survey / analysis (2)、 The product design / Product solution definition (3)、 Project solution definition (4)、 Product development process management

2、Project Manager/ project manager —B1 role

The company is in some technical fields , such as “ Face recognition ”、“ Vehicle identification ”、“ Voice to text ”、“ object detection ” And other technical directions have achieved relatively good accuracy , Basically reach or exceed SOTA（State-of-the-art） level , It has the possibility of commercial application , There will also be some industry customers to inquire about the application of relevant technologies , More firm confidence in commercialization .  

background	When we complete product development , Or the product may be semi-finished or just MVP, After successful sales to government and enterprise customers , Then the project needs to be officially launched . The product manager and the project manager are the product and project managers respectively “ The general manager ”, It's implementation technology → product → The core backbone of landing delivery . There are parts that are completed independently , There are also parts to be completed jointly .
duty	The project manager needs to complete the management 、 Coordinate 、 control 、 Delivery and other work within the project life cycle . however , in fact , At the practical level , There is a lot of “ Support products with projects ” The situation of . (1)、 To develop project plan ： Fully understand the goal , Make an implementation plan ; (2)、 Complete the delivery of the scheduled objectives of the project as planned ： Coordinate internal and external resources , Follow up the implementation progress , control risk , Responsible for end-to-end delivery ; (3)、 Weekly project meeting ： Organize various review meetings and regular meetings of the project (4)、 Coordinate and follow up the landing ： Implementation of internal and external cooperation
Main work outputs	(1)、 Project documentation (2)、 Project solutions (3)、 Project delivery plan
Excellent qualities	(1)、 Product sense insight ： Especially the projects related to artificial intelligence , In addition to the management ability, the project manager , The product manager's “ Product sense ” And “ insight ”.
Workflow	Project solutions Project implementation plan 、 Project delivery （ The implementation of ） programme Demand survey project management Balance project management “ Iron triangle ”

How to quickly realize commercialized products ？

that , How to realize commercial products ？ For technology driven companies , The probability is that there is no knowledge reserve of relevant complete products and industry solutions , therefore , We need to run fast in the market of government and enterprise customers , go “ shortcut ”, Quickly form products 、 Solution and product matrix . The basic process is as follows ：

(1)、 Refer to peer solutions as the basis of this scheme	According to customer needs , In industry benchmarking , Look for competitors' existing products and solutions 、 Product architecture 、 Technical architecture and core functions .
(2)、 Business and technical teams define solutions	The route of the solution , You have to have Have enough competitiveness .
(3)、 Business and product team design PPT programme	solution-based PPT, You have to have Have a high standard .
(4)、 Finish quickly Demo and MVP Development of	Based on the company's technology accumulation 、 Historical case 、 Existing products , Combine the core needs of users and their pain points , Finish quickly Demo and MVP Development of .
(5)、 Build a requirements team	After winning the project through bidding , Then set up the project team . In the project team , The product manager and some members of the solution team will be specially arranged , form “ Requirements team ”.
(6)、 Summarize product requirements	be based on The contents of the contract require , combination On site user demand survey , Get your first-hand needs , At the same time, analyze the product functions of competitors in the industry . The project manager leads “ Requirements team ”, Completion can meet the project delivery , And it is general enough in the industry “ Product needs ”— This is an ideal state , In the actual operation, there is pain , affair .
(7)、 Determine the architecture and technical roadmap	The requirements team and the architecture team discuss and determine three major events ： Product base / Universal / Support function , Product features and functions , Project functions , Based on this understanding , Start Design architecture level and select technical route ; emphasize , The project is time bound , Not to pursue “ Ideal ” Ignoring progress .
(8)、 Formal development of the project ( Branch / Merge )	The R & D team quickly completed in master（Dev） Branch , Complete the first part of basic and general function development ; then , Branch out “pd-xx” Develop the product features and functions required by the project , Tests merged into master after ; Branch out “pj-xx” Develop project functions . Before the project development is completed , The development of the product branch still serves the project , To quickly merge .
(9)、 Improve and expand demand	Given that the needs of a single customer must have certain one sidedness and limitations , While doing project development , The requirements team needs to further Research industry demand , Quickly update and improve the requirements of the product .
(10)、 Build complete project and product versions	ideally , After project delivery , In our version control system, there is a complete project version and a basically mature product version , That's it Support products with projects , At the same time, the R & D cycle and cost are minimized .

3、Business Consultant/ Business consultants ( coordination PM)— Role concept —A2 role

background	Business consultants are more like “ role ” Concept , Not all companies set up full-time positions . In general , The role of business consultant is played by two types of personnel . The first category , Business professionals , stay Rich working experience in the target industry , Participate in business consulting as a business expert ; The second category , New people in society , Aspire to be a business consultant , Generally, I am familiar with module functions 、 Participate in business demand research and begin to cultivate , Depending on personal development intention , The product manager will be informed later 、 The project manager or sales direction is transferred .
duty	(1)、 Communicate business and market needs (2)、 Organize industry solutions ： Review the existing scheme 、 feedback (3)、 coordination PM Design requirements document ： Cooperate with product manager / The project manager completes the product / Project demand survey , And form corresponding requirements documents ; Design the pre-sales consulting scheme PPT： Cooperate with the sales manager , Complete the pre-sales consultation and products of the project / Introduction of the plan , And form a pre-sales consulting scheme /PPT; Suggestions on product direction ： industry / Competing goods analysis , Understand trends and trends , Give product direction suggestions ;
Main work outputs	(1)、 Demand Research Report →PRD (2)、 Solution
Excellent qualities
Workflow	Demand survey Solution Business analysis

 

4. Data Scientist/ Data scientist —C1 role

background	Data scientist is a new job in recent years , Different companies 、 The industry sector is for “Data Scientist” The definition of this role may vary , But as an interdisciplinary position , Generally speaking, it has the following work contents ：
duty	(1)、 Participate in project pre-sales consultation (2)、 Digital innovation and solution development and implementation (3)、 Mining data value ： Digital drive brings value to customers , Innovatively meet business needs . Data processing and data analysis , Business value analysis and reporting . Conduct complex modeling and discover the business value and significance of the data . (4)、 Guiding modeling ： Apply various types of statistical modeling 、 Machine learning and deep learning methods solve practical problems . Mainly involved in ML The job of , Guide business modeling through data analysis 、 model training .
Main work outputs	(1)、 Data analysis report (2)、Models
Excellent qualities	(1)、 Data sensitivity ： (2)、 Summarizing ability ： Be able to summarize and preach the value of excavation . (3)、 Presentation skills ： Have good written 、 Oral communication 、 Reporting capacity
Workflow

5. ML Researcher/ Machine learning researcher —C2 role

background	The main task of researchers is more forward-looking exploration and research , Track industry technology trends , Solving modeling problems in new scenarios , To optimize the accuracy and performance of the algorithm 、 Promote the application and implementation of the algorithm .
duty	(1)、 Technology tracking ： Prospective research and tracking technology trends , Propose innovative idea; (2)、 Feasibility study ： Feasibility demonstration of algorithm implementation 、 Project feasibility demonstration (3)、 Architecture design ： Reasoning system architecture and design (4)、 model training 、 Algorithm tuning
Main work outputs	(1)、Models (2)、Paper (3)、 analysis / assessment report
Excellent qualities
Workflow

6. Data Engineer/ Data Engineer —C3 role

background	The main task of the data engineer is to maintain the data pipelines, complete data management pipeline Data extraction 、 polymerization 、 cleaning 、 Storage and automated pipeline monitoring , Make sure that the following ML Working data availability .
duty	(1)、 Data storage design in the development project 、 Development and performance optimization work (2)、 data polymerization / Storage → data ” modeling ”： The data model here refers to... In the database table (3)、 Count → data cleaning → data mining (4)、 data monitor
Main work outputs	(1)、 database / Data warehouse (2)、sql Access related codes
Excellent qualities
Workflow

7. ML Engineer/ Machine learning Engineer —C4 role

background	Machine learning engineers generally build on the work of data scientists and researchers , Further training and tuning of the model , To match items / Technical indicators set for the product , And complete the deployment of reasoning system and model and the integration with business software system .
duty	(1)、 Solve the need ： Using data mining / machine learning / Deep learning and other related algorithms , Address core business needs (2)、 Algorithm selection → Project realization → Optimization and innovation ： model training 、 Algorithm tuning (3)、 Reasoning system deployment ： Model deployment 、 Reasoning system interface development (4)、 Software system integration ： Inference system , Integration with software products (5)、 Build machine learning tools and code bases
Main work outputs	(1)、Models (2)、 Inference system (3)、 Software integration
Excellent qualities	(1)、 Coding ability ： For machine learning Engineers , Have certain code implementation ability and master a mainstream back-end development language （Java、C#、.NET Core、C++、Go etc. ） It's necessary .
Workflow

 

8. DevOps Engineer/DevOps The engineer —C5 role

background	DevOps Has been widely recognized and practiced , The relevant tool chain has changed from open source free to closed source commercial 、 From privatisation to deployment SaaS service , Extremely rich and perfect , When there are too many choices , How to choose and apply has become a new problem . Choose the right tools to automate the process ,DevOps The requirements for knowledge points and skill stacks involved are relatively high , therefore DevOps The engineer was born .
duty	(1)、 structure DevOps technological process ： And analyze 、 Optimize DevOps practice (2)、 Build infrastructure platform ：DevOps pipeline Selection of tool chain , Deploy 、 Maintenance and monitoring (3)、 Establish continuous CI/CD Environmental Science ： Accelerate the software development and deployment process , Risk aversion
Main work outputs	(1)、DevOps Cultural guidance (2)、DevOps Infrastructure platform (3)、 Product deployment in user privatization environment
Excellent qualities
Workflow

9、Software Develop Engineer/ Software Development The engineer —D1 role

As mentioned above , This paper mainly discusses the task of machine learning method , So for Traditional Software Engineering The different roles involved in this section , such as ： Architects 、UI、UE、 front end 、 Back end 、 Mobile terminal and function test 、 Automated testing, etc , All merged into “Software Develop Engineer（SDE）” A uniform reference to .

background	For software development engineers , actually , As a product or project delivery system , machine learning / Deep learning model Part of the time is not high , It is more inclined to the architecture of traditional software engineering 、 Tools 、 data 、 Code and other contents , And this part is SDE The team is better at and familiar with .
duty	(1)、 Architecture design (2)、 Interface design ：APIs、MQ、JSON The contents of the agreement shall be agreed in advance 、 Design ; (3)、 Software ( front / Back end ) Development 、 test
Main work outputs	(1)、 And DL Reasoning system integration (2)、 The software product
Excellent qualities
Workflow	about SDE Part of the work content , Don't introduce too much , Everyone is very clear about the division of labor 、 Work content and deliverables . For the part involving deep learning , Here are some suggestions ： (1)、 about CV Class may Web The page needs to show the real-time video stream and callout box , Make an appointment in advance ： A： Transfer coordinates to front-end drawing ; B： Reasoning platform OpenCV Draw a callout box , And whether to transcode to video stream ; (2)、 It is also necessary to evaluate the front-end display effect after frame extraction , Whether it meets the needs ; (5)、 Business side to DL Servin End call requirements 、 Task and other interface design ;

10. Delivery Engineer/ Delivery Engineer —D2 role

background	In the traditional field of software delivery, there is this kind of “ Consensus ”—— Three point software , Seven points of implementation . The delivery phase is for software projects , It's very 、 very 、 Very important , Never underestimate . This work , By no means arrange a project manager , Take a few fresh graduates who are willing to travel 、 A child who can bear hardships can finish it smoothly . Then consider the projects brought about by the deep learning part The complexity of technical concepts 、 Unpredictability of results 、 Unexplainability , Will make the project Delivery is harder , It is more necessary to have a professional team to complete the delivery work .
duty	Implementation and delivery of machine learning projects , Under the leadership of the project manager , A team of delivery engineers with different skills and abilities , Complete project delivery and implementation . (1)、 Project needs research (2)、 Field test 、 The implementation of (3)、 Complete the project deliver 、 check before acceptance (4)、 Complete customer training of software system 、 Technical support ： Solve the customer's problems
Main work outputs	(1)、 Project demand research report (2)、 Project plan 、 Implementation plan (3)、 Project process document (4)、 Product development delivery documents
Excellent qualities	(1)、 EQ ： communicate 、 Ability to coordinate and manage projects 、 Ability to communicate and judge demands , (2)、 Knowledge reserve ： Industry understanding 、 Process cognition and familiarity with industry-specific terms (3)、 Copywriting skills ： Ability to write and report official documents (4)、 Technical ability ： System architecture evaluation and code implementation evaluation capability , machine 、 The Internet 、 Engineering and technical capabilities for system integration such as storage (5)、 Comprehensive ability ： Meeting ability 、 Risk management capability , Ability to locate and troubleshoot problems ; Pressure resistance and psychological quality ; Generic cabling 、 Construction management and communication with workers ; Versatility and flexibility
Workflow	(1)、 Project needs research (2)、 Develop project delivery plan 、 Strategy (3)、 System deployment and field test (4)、 Train users to use (5)、 Monitoring after online

Reference article
AI Project implementation guide --3. The machine learning project team consists of - You know