How to Transition into AI: Real Stories, Tools, and a Practical Roadmap

Preface

“When a person gets tired of learning technology, they will also get tired of the IT industry, because only continuous learning brings the rewards of IT, including money.”

The IT field demands constant learning; AI is no exception. The author, inspired by the book *Londoners*, writes this piece to discuss AI talent growth.

Participants

Mike Zhou: Graduated from Zhejiang University in 2004, started programming at 15 with BASIC, later wrote C and JSP, moved to distributed software, big data, and spent the last four years on AI platform product development.

Mr Qiu: Master's in algorithms, five years of experience; studied AI and electronics at Hangzhou Dianzi University, worked on computer vision, algorithm optimization, and training framework improvements.

Hannah: B.E. in Electrical Engineering from Manchester, M.Sc. in Data Science from UCL; two years at Huawei working on algorithm research, platform development, and POC projects.

Doctor Zeng: Ph.D. from a top Chinese university, Huawei Special Offer in 2018; extensive experience in information systems, AI projects, NLP, and knowledge graphs; now chief expert of an algorithm modeling team.

Fan: 20 years in ICT, experience in signal processing, networking, OS architecture, big‑data analysis, and AI product planning; currently chief product management expert for Huawei Cloud ModelArts.

Reading Recommendations

The author suggests starting with Nick’s *A Short History of Artificial Intelligence* for a broad overview, then the textbook‑style *Artificial Intelligence (2nd Edition)* by Stephen Lucci and Danny Kopec. Depending on goals, further reading may include Zhou Zhihua’s *Machine Learning*, the *Deep Learning* book by Ian Goodfellow et al., *Hands‑On Deep Learning*, and practical guides for TensorFlow and PyTorch.

Hands‑On with PyCharm and ModelArts

To quickly develop and train models on Huawei Cloud, the author uses the PyCharm‑ToolKit to connect a local PyCharm IDE with ModelArts. The steps include:

Install PyCharm Professional 2019.2 (the Toolkit only supports this version).

Download the Toolkit zip (PyCharm‑ToolKit‑PC‑2019.2‑HEC‑1.3.0.zip) and install it via the PyCharm Plugins menu.

Configure OBS credentials in Huawei Cloud and import them into PyCharm.

Create a ModelArts training job (e.g., MNIST digit recognition using MXNet), monitor logs, and retrieve the trained model from OBS.

Key screenshots illustrate the download links, plugin installation, credential entry, and training job execution.

AI Career Path

The author proposes three stages for a successful AI transition:

Get hands‑on: Run open‑source projects such as Darknet YOLOv3 to understand data flow and GPU‑framework interaction.

Master principles: Study the evolution of YOLO (v1‑v4) and later explore R‑CNN for higher recall.

Deploy inference: Optimize models (FP16/int8), use accelerators like NVIDIA TensorRT or Huawei Da Vinci chips, and integrate the model into production systems.

Business‑Driven AI Project Workflow

For a real‑world AI project (e.g., credit‑card fraud detection), the workflow includes:

Business understanding: Define the problem, resources, and goals with domain experts.

Data acquisition & cleaning: Collect relevant data, handle redundancy, missing values, outliers, and inconsistencies using techniques such as deletion, imputation, and clustering.

Data splitting: Create training, validation, and test sets; address class imbalance for fraud detection.

Feature engineering: Spend the majority of effort on extracting and refining features; iterative improvement is essential.

Model selection & training: Choose algorithms (e.g., LightGBM, XGBoost, Random Forest) and tune hyper‑parameters.

Inference & evaluation: Run the model on test data, evaluate with metrics such as precision, recall, F1, and adjust as needed.

Deployment: Deploy the model to the business system, monitor performance, and continuously retrain with new data.

Several screenshots show the ModelArts Miner evaluation UI, data split operator, and end‑to‑end workflow.

Conclusion

The author encourages continuous learning, leveraging platform resources, and participating in community courses such as Huawei Cloud AI Bootcamp. By following the outlined reading list, hands‑on practice, and business‑centric workflow, professionals can successfully transition into AI development.