What DataArt Looks for in Machine Learning Engineers: Expected Skills for Each Level and Open Positions

DataArt's machine learning engineers work across finance, healthcare, retail, travel, media, and education — some of the most data-intensive industries there are. They also build and contribute to DataArt's own AI platforms, including Artisyn, a secure AI-native delivery platform, and AILA, an agentic development system that has cut software delivery time by 74%.

What the work actually looks like

Because DataArt is a software engineering services firm, ML engineers here work across domains and project types rather than spending years on a single product. That variety accelerates growth — and it also means that engineers who combine technical depth with clear communication tend to do particularly well. If you enjoy both solving hard problems and explaining them, you'll find a natural home here.

One important note before diving in: at every level — Junior, Middle, and Senior — you're not expected to know all of this. Roughly 60–70% coverage of the scope below is a good orientation point. The lists describe the full picture of what's relevant at each level, not a checklist you need to pass. Gaps in specific areas are normal and expected to fill in on the job.

Junior ML Engineer

At the junior level, we're looking for solid fundamentals across the core ML toolkit: linear and logistic regression, decision tree algorithms, boosting and bagging, and SVMs. You should understand overfitting and underfitting, know how to tune hyperparameters, visualize data, and measure model performance on classical classification and regression tasks.

On the neural network side, you'll need to understand metrics and backpropagation. Exposure to basic NLP concepts (Word2Vec, GloVe, FastText embeddings, Transformers finetuning) and CV tasks (image classification, object detection, classical image segmentation, OCR) is a plus but not required at this stage.

For programming and frameworks, Python is expected, along with PyTorch, TensorFlow, XGBoost/LightGBM/CatBoost, sklearn, pandas, numpy, OpenCV, Flask, and SQLAlchemy/Psycopg2. API development and Docker are part of the production baseline from the start, as is production-ready code: PEP8, conventional commits, clean structure. Some hands-on exposure to LLM APIs (OpenAI, Anthropic, Gemini) is helpful but not required.

Cloud exposure is not required at Junior — any familiarity with AWS SageMaker/Lex or Azure OpenAI service is a plus. Broader cloud breadth comes at Middle.

One thing that stands out at this stage: we expect production-quality code from day one. If you've mostly worked in notebooks, it's worth spending time on that before applying — it'll make a real difference in how the technical interview goes.

Middle ML Engineer

The jump from Junior to Middle is largely about depth and independence. In NLP, finetuning transformers becomes a core skill alongside understanding the mechanics behind it — embeddings training, vanishing and exploding gradients, and MLP/CNN/LSTM architectures. Training transformers from scratch comes in at Strong Middle and Senior. In computer vision, the scope expands to deep segmentation methods (U-Net and other deep neural networks), ResNet/YOLO/FaceNet training, augmentations, OCR training, and human-in-the-loop labeling.

Time series analysis becomes part of your toolkit at this level: Prophet, GluonTS, LSTM, CNNs, cross-validation techniques, decomposition and stationarity, ARIMA-family and exponential smoothing models, and time series visualization.

On the engineering side, the expectations grow considerably: Bash/PowerShell scripting, distributed computing, CI/CD, pipeline automation, model monitoring, and diagram creation. Async programming for AI pipelines and streaming responses (SSE, WebSockets) start to show up as the work increasingly involves LLM-based services. The framework stack expands to include Horovod, FastAPI/Aiohttp, Flask, and SQLAlchemy/Psycopg2.

GenAI exposure is welcome but not required at Middle: working with LLM APIs (OpenAI, Anthropic, Gemini) and open-source LLMs from Huggingface, understanding their limitations and basic hallucination mitigation, and putting together quick demos and POCs with Streamlit/Gradio or modern vibecoding tools (Cursor, Claude Code, Copilot, and similar) all strengthen a Middle profile.

On the cloud side, AWS is where the bar is at Middle: meaningful experience across EC2, ECS/Kubernetes, SageMaker, ECR, Lambda, Lex, Textract, Rekognition, and Bedrock. Working knowledge of GCP (Vertex AI, Cloud Functions, GenAI Studio/Gemini API) or Azure (OpenAI service, Conversational AI/AI Search, Cognitive Services) is a plus rather than a requirement at this level.

The soft skills piece becomes increasingly important at this level — though formally still a plus rather than a requirement. DataArt values Middle engineers who understand the business context behind their work and can communicate it clearly to clients. Storytelling and domain understanding tend to be what makes the difference on projects.

Senior ML Engineer

Senior ML engineers at DataArt are technically comprehensive and commercially confident. On the technical side, that means RLHF, Transformers for time series, distributed model training, TensorFlow Extended/Serving, H2O, and full-stack cloud deployment — including GCP Cloud Engine and Cloud ML/Vertex AI AutoML, and Azure Functions and App Insights alongside the full AWS, GCP, and Azure stacks.

The Senior bar also reflects how much the field has shifted toward GenAI. Senior engineers are expected to be fluent with LLM API usage and open-source LLMs, prompt engineering (chain-of-thought, ReAct, few-shot/zero-shot, prompt optimization), and the building blocks of RAG systems — vector databases, embedding strategy, chunking, advanced RAG techniques, and RAG evaluation. Agents and tool/function calling, MCP, and frameworks like LlamaIndex, LangGraph, CrewAI, and Semantic Kernel are part of the picture. Knowing when to reach for classical ML versus LLMs — particularly on tabular data or in regulated industries — is part of the judgement we look for. Image and video generation pipelines come up on some projects but aren't required.

Security and responsible AI are part of how we expect Seniors to think about production systems — data privacy, prompt injection prevention, guardrails and content filtering, AI security architecture, and compliance (GDPR, SOC2, HIPAA, EU AI Act). Broader system-design and AI-ops topics — LLM architecture patterns (RAG, agents, multi-agent), embedding strategy and vector store architecture, model versioning and A/B testing, AgentOps/LLMOps, observability, and AI risk assessment — strengthen the profile but are not hard requirements.

What's distinctive about the Senior role at DataArt is the pre-sales dimension. Senior engineers often participate in pitching and scoping new work — translating complex ML capabilities into business value for prospective clients, presenting AI solutions to non-technical stakeholders, and leading discovery workshops. Helping with solution design and project estimation across compute, tokens, and infrastructure costs is a plus but not required. Many engineers find this one of the most energizing parts of working at a services firm: you get to help shape what gets built, not just build it.

What the technical interview looks like for ML roles

The technical assessment at DataArt has two parts: theoretical knowledge and real-project thinking. Expect questions that probe your understanding of core ML concepts — not just "have you used XGBoost" but "when would you choose it and why" — along with how you think about the full ML lifecycle from data preparation through deployment and model monitoring.

For more senior candidates, expect to walk through past project decisions: what tradeoffs you made, how you worked with stakeholders, and what you'd do differently. There's no algorithm puzzle gauntlet — the goal is to understand how you think and communicate, which is something you can prepare for with real project stories.

Before the technical round, there's an HR conversation and an English communication evaluation — DataArt's teams and clients are global, and strong English is a genuine requirement at all levels. The full interview process is outlined here.