AI Engineering
Machine Learning Engineer
A model that works on a laptop and a model that works for millions of users are two very different artifacts, and machine learning engineers live in the gap between them. The role exists to take research-grade ML and turn it into reliable production systems, which means caring about latency, retraining pipelines, and what happens when the data distribution shifts at three in the morning.
Students grow into this through hands-on work with PyTorch or TensorFlow plus enough software engineering discipline to run real CI/CD. Tools like AWS SageMaker become part of the workflow.
Strong ML engineers can talk shop with data scientists on one side and platform engineers on the other, and that bilingual quality is often what gets them hired.
- CodeSeniorNew
Train a Manipulation Policy for Bin Picking with Imitation Learning
You receive a dataset of 500 teleop trajectories on the in-distribution part plus a held-out simulation environment with a never-seen part. Train an imitation-learning policy (D…
- Imitation Learning
- Manipulation
- Diffusion Policy
Advanced Robotics - ResearchSeniorNew
Inductive Logic Programming for a Fraud-Rule Discovery Pilot
You receive a labeled fraud dataset (around 25,000 transactions, around 4% positive class), a feature schema (28 features including device, geo, behavioral history), and a basel…
- Inductive Logic Programming
- Symbolic Ai
- Rule Learning
Fuzzy Logic, Knowledge Representation, and Symbolic Reasoning - ResearchSeniorNew
Price American Options with a Deep Hedging Notebook
Simulate price paths for a single underlying (geometric Brownian motion is fine as a baseline; bonus for stochastic volatility). Implement Longstaff-Schwartz Monte Carlo as the …
- Deep Learning
- Stochastic Modeling
- Derivatives Pricing
AI and Quantitative Finance - DesignSeniorNew
Design a Distributed Training Job for a 13B-Parameter Model
Decide whether to use Fully Sharded Data Parallel (FSDP), Tensor Parallelism, Pipeline Parallelism, or a hybrid; justify against the 13B-param + 32-H100 setup. Calculate memory …
- Distributed Training
- Fsdp
- Pytorch Or Tensorflow
Machine Learning Systems Practice your coursework on real scenarios.
Every challenge is shaped from real-world context — not generic exercises. The work mirrors what your degree prepares you for.
Why Ewance
- ResearchSeniorNew
Solve a POMDP for a Healthtech Diagnostic Pathway
You receive a simplified pathway: 5 possible underlying conditions, 8 possible diagnostic tests each with documented sensitivity and specificity, and an outcome payoff matrix fr…
- Pomdp Modeling
- Belief States
- Approximate Solvers
Decision Making Under Uncertainty - ResearchSeniorNew
Structure Learning for a Causal Network in Fintech Risk
You receive the 60-signal dataset and a short interview summary of risk analysts' beliefs about which signals influence which. Use a hill-climbing structure-learning algorithm w…
- Structure Learning
- Bayesian Networks
- Causal Modeling
Probabilistic Graphical Models - CodeSeniorNew
Video Action Recognition for a Retail Loss-Prevention Startup
Use a public action-recognition dataset (UCF101 + a small curated retail-action subset; the latter is provided synthetic or you can label 50 short clips). Fine-tune a small back…
- Video Understanding
- Action Recognition
- Transfer Learning
Computer Vision - ResearchSeniorNew
Diffusion-Policy Imitation for Bimanual Cooking Tasks
You receive 300 teleoperated demonstrations of a bimanual pour-and-stir task in a Robomimic-style simulator, deliberately including 2 valid solution modes per task (left-pour-ri…
- Diffusion Policies
- Imitation Learning
- Multimodal Action Distributions
Robot Learning - Browse challenges
Explore role
Product Manager
Ship product that solves real user problems. Combine user research, prototyping, and stakeholder alignment to turn ambiguous briefs into measurable wins — the role at the centre of modern software teams.
- ResearchSeniorNew
Open-Vocabulary Segmentation Benchmark for a Robotics R&D Lab
Use a curated 200-image household scene set (publicly-available HM3D renderings or COCO + a handful of household prompts). Benchmark 3 open-vocabulary segmentation models: SAM +…
- Open Vocabulary Segmentation
- Vision Language Models
- Benchmarking
Computer Vision - ResearchSeniorNew
Pretrain a Small Vision Transformer with Self-Supervised Learning
You receive 80,000 unlabeled 224x224 histology tiles plus 4,000 labeled tiles split into train/val/test. Pretrain a ViT-Small using a self-supervised method of your choice (DINO…
- Supervised Learning
- Vision Transformers
- Pytorch Or Tensorflow
Advanced Deep Learning - CodeSeniorNew
Triage Brain-CT Stroke Detector with Calibrated Uncertainty
You receive a curated public head-CT dataset (about 2,800 scans, slice-level labels for hemorrhagic stroke) and a held-out 600-scan hospital cohort. Train a 3D CNN or 2.5D slice…
- Medical Imaging
- Neural Networks
- Uncertainty Quantification
Machine Learning for Imaging and Medical Image Analysis - ResearchSeniorNew
DPO Preference-Tune a Code Assistant for Style Compliance
You receive a 7B coding base model, a client's published code-style guide (Python, around 80 pages), and a generated preference dataset (4,000 pairs of code snippets where one m…
- Dpo
- Preference Optimization
- Fine Tuning
Fine-Tuning Large Language Models Build a verifiable portfolio.
Submissions become evidence. Reviewers with shipping experience score against a rubric; the result becomes a credential anyone can verify.
Why Ewance
- ResearchSeniorNew
Train a Small Diffusion Model for Synthetic Defect Generation
You receive 2,000 labeled defect images and 18,000 clean weld images. Train a small class-conditional latent diffusion model on the defect images (Hugging Face diffusers is fine…
- Generative Perception
- Diffusion Models
- Data Augmentation
Machine Perception - ResearchSeniorNew
Reproduce a Mechanistic Interpretability Result on a Small Transformer
Pick a published mechanistic-interpretability paper that operates on a small (under 1 billion parameter) open-source transformer (e.g., GPT-2 small, Pythia 70M). Set up the envi…
- Mechanistic Interpretability
- Transformer Internals
- Pytorch Or Tensorflow
AI Safety and Alignment - ResearchSeniorNew
Curriculum RL for a Simulated Drone Inspection Task
You receive a PyBullet-based wind-turbine inspection simulator with parameterizable wind, blade orientation, and sensor noise. Design a 3-stage curriculum: (1) hover near a stat…
- Ppo
- Curriculum Learning
- Deep Rl
Reinforcement Learning - ResearchSeniorNew
Trajectory Prediction Model for Urban Robotaxis
Use the Argoverse 2 motion-forecasting dataset (open access). Train an LSTM baseline + a transformer challenger (e.g., a small Wayformer or HiVT). Evaluate on minADE/minFDE (min…
- Trajectory Prediction
- Transformer Models
- Evaluation
AI for Autonomous Vehicles - CodeSeniorNew
Profile and Cut Inference Cost on a Recommender at Scale
You receive (1) a frozen ONNX export of the production model, (2) a sample request trace of 24 hours at 1% sampling, and (3) a single A100-class GPU sandbox. Profile with NVIDIA…
- Gpu Profiling
- Model Quantization
- Inference Optimization
Machine Learning Systems - ResearchSeniorNew
Quantify Sim-to-Real Gap for a Warehouse Manipulation Policy
You receive a trained pick-and-place policy (PyTorch), the simulation env (Isaac Lab), and access to a real-arm rig (or recorded teleop episodes if hardware is unavailable). Def…
- Sim To Real
- Manipulation
- Experimental Design
Robot Perception and Autonomy - CodeSeniorNew
Survival-Analysis Risk Model for an Oncology Decision-Support Pilot
You receive a curated public colorectal cancer cohort (about 9,000 patients, demographics, stage, grade, comorbidities, baseline labs, censored survival times). Fit (1) a Cox pr…
- Survival Analysis
- Risk Stratification
- Model Calibration
Machine Learning for Healthcare and Biomedicine - CodeSeniorNew
Coordinate a Fleet of Warehouse Robots
Implement a simulated warehouse grid with 80 robots solving a pick-and-deliver workload. Design a decentralized coordination protocol (recommend a contract-net or auction-based …
- Multi Agent Coordination
- Decentralized Algorithms
- Simulation
Multi-Agent Systems - ResearchSeniorNew
SAT-Based Planner for Smart-Grid Demand Response
Encode the dispatch problem (which customers to curtail by how much, respecting per-customer contractual caps and grid-cell totals) as a SAT or MaxSAT instance. Solve 50 histori…
- Sat Based Planning
- Constraint Encoding
- Benchmarking
Automated Planning - ResearchSeniorNew
Long-Context QA Evaluation Benchmark for Legal Memoranda
You receive 25 anonymized legal memoranda (50-90 pages each) and 100 QA pairs whose answers are deliberately spread across the documents (25 in pages 1-20, 25 in pages 20-40, 25…
- Long Context Qa
- Benchmark Design
- Model Evaluation
Question Answering and Conversational Systems - ResearchSeniorNew
Self-Supervised Pretraining for a Pathology Foundation Vendor
You receive a public pathology dataset (about 80,000 unlabeled whole-slide-image patches plus a labeled 8,000-patch subtype-classification subset across 4 classes). Pretrain a R…
- Supervised Learning
- Medical Imaging
- Transfer Learning
Machine Learning for Imaging and Medical Image Analysis - ResearchSeniorNew
Design a Distributed-Training Strategy for a Mid-Sized LLM
You will write a 5-page design memo that picks a parallelism strategy for fine-tuning a 13B model on 32 H100 GPUs, with a tokens-per-second estimate, a memory-per-GPU calculatio…
- Distributed Training
- Parallelism Strategies
- LLM Training
Machine Learning at Scale - CodeSeniorNew
Train a Reinforcement-Learning Policy for Drone Obstacle Avoidance
You receive a custom Gymnasium drone-flight environment (provided), a baseline hand-engineered controller, and a target evaluation suite covering 4 obstacle densities. Train a P…
- Reinforcement Learning
- Ppo
- Robotics Simulation
Advanced Robotics - CodeSeniorNew
PPO Alignment Loop with a Pretrained Reward Model
You receive a small open-weights base model (around 7B), a previously trained reward model, and 5,000 prompts (no responses) for PPO rollouts. Run PPO with TRL's PPOTrainer for …
- Rlhf
- Ppo
- Reward Hacking
Machine Learning from Human Preferences (RLHF and Alignment)
How it works
From brief to credential, in six steps.
Step 01
Browse challenges aligned to your studies.
Step 02
Accept the one that fits your goals.
Step 03
Work through it with AI Copilot guidance.
Step 04
Submit for structured evaluation.
Step 05
Earn a verified credential.
Step 06
Add it to LinkedIn with one click.
Related roles you may want to explore
Browse all roles →AI Engineering
AI Engineer
Between a promising research paper and a feature people actually use sits a long, unglamorous bridge — and AI engineers build it. The job is taking models that work in notebooks and shaping them into systems that hold up under real traffic, real costs, and real users with messy questions. Good work here looks like a retrieval pipeline that gets answers right ninety-something percent of the time, with evaluation harnesses catching regressions before they ship. Students grow into this role by treating Python and PyTorch as instruments rather than checkboxes, then learning how to reason about latency, evaluation, and cost together. If you enjoy stitching ideas into running software, this path will feel like home.
AI Engineering
Computer Vision Engineer
Teaching a machine to see is harder than it sounds and more interesting than it looks. Computer vision engineers shape the systems that read documents, navigate self-driving cars, screen medical images, and answer questions about photos. The role mixes the math of multi-view geometry with the engineering grind of getting models small and fast enough to run where they're needed — sometimes on a phone, sometimes on a robot. Good work here looks like a pipeline that holds up in real lighting, real motion, and real failure modes. Students grow into this path by getting hands-on with OpenCV and PyTorch early, then learning the harder craft of optimizing models without quietly destroying their accuracy.
AI Engineering
MLOps Engineer
Models in production fail in stranger ways than models in notebooks ever could. The MLOps engineer is the person who anticipates those failures and builds the scaffolding that makes machine learning survive contact with real users. Think feature stores that stay consistent between training and serving, deployment pipelines through MLflow that make rollbacks boring, and observability that catches drift before stakeholders notice. The work sits at the intersection of platform engineering and data science, and rewards people who like building tools other engineers will rely on. A student grows into this role by getting comfortable with Kubernetes early and developing taste for what a healthy ML system actually looks like under load.
AI Engineering
NLP Engineer
Language is messy. People misspell, contradict themselves, ask the same thing five different ways, and expect a machine to understand. NLP engineers build the systems that try. The role spans classical text processing in spaCy, modern retrieval-augmented architectures stitched together with LangChain, and the constant judgment calls about when to fine-tune, when to prompt, and when to fall back to rules. It rewards people who love both linguistics and systems thinking. Students grow into it through small projects — a question-answering bot over their notes, a classifier for their inbox — that surface the real failure modes of language models. Good NLP engineers obsess over evaluation as much as architecture.
AI Engineering
Prompt Engineer
Writing instructions for a model is a strange new craft. The words you choose, their order, the examples you include — all shape what a multi-billion-parameter system actually does next. Prompt engineers treat this as a real engineering discipline: versioning prompts in tools like PromptLayer, running evaluations across thousands of test cases, optimizing for cost and latency in production, and collaborating with domain experts to encode their judgment in text. The role is new enough that students often help define it on the job. Growing into it means building intuition for how models fail, when to fine-tune instead, and how to write specs precise enough to ship. Good prompt engineers measure everything and trust vibes only as a starting point.
Industry teams behind a decade of practitioner briefs
Hiring from this pool?
Sponsor a challenge and meet candidates through actual work.
Industry teams can shape briefs around the skills they hire for, then evaluate students on rubric-scored deliverables — not resumes.
Skills and disciplines shown on this page are derived from the Ewance challenge catalogue. When the median annual salary is available for this role via Adzuna, it will be shown above with the sample size and country.
Portrait: Photo by Yevgeniy KHVAN on Unsplash.