Evaluation
If you like applying Evaluation, every challenge here gives you a chance to practice it on a real industry brief.
- CodeIntermediateNew
Semantic Parser for an Enterprise Analytics Assistant
Define a small typed query language (filter, aggregate, group_by, time_range, metric). Curate or write 200 training examples covering the controlled subset and 50 held-out test …
- Semantic Parsing
- Grammar Design
- Transformer Models
Computational Semantics - CodeIntermediateNew
DPO Fine-Tune for a Domain-Specific Writing Assistant
You receive a base instruction-tuned model checkpoint plus 2,500 preference pairs from editorial reviews (each pair: two grant-application paragraphs, the editor-preferred winne…
- Dpo
- Preference Learning
- Model Finetuning
Machine Learning from Human Preferences (RLHF and Alignment) - CodeIntermediateNew
Fuse LiDAR and Camera for an Autonomous Yard Truck
You receive 6 hours of synced LiDAR + 4-camera ring data from yard operations, with 3D bounding-box labels for pedestrians, forklifts, and containers. Build a late-fusion module…
- Sensor Fusion
- Lidar Perception
- Object Detection
Open coursework - ResearchIntermediateNew
Visual Question Answering for a Pediatric Radiology Workflow
You receive ~8,000 publicly available pediatric chest X-rays with structured findings labels (anonymized; no PHI access required). Build a VQA pipeline that maps a (image, quest…
- Vision Language Models
- Visual Question Answering
- Lora Finetuning
Open coursework 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
- ResearchIntermediateNew
Fine-Tune a Vision-Language Model for Image Captioning
Take BLIP-2 or LLaVA-1.6 as the base. Fine-tune (LoRA is fine) on a 4,000-image accessibility-curated dataset where each image has a useful caption written by a low-vision-exper…
- Vision Language Models
- Lora Fine Tuning
- Pytorch
Multimodal Machine Learning - CodeIntermediateNew
Segment Cells from Microscopy Images for a Pharma-AI Discovery Lab
You receive 3,500 microscopy images with pixel-level cell masks plus a 200-image hold-out set re-annotated by two biologists for inter-annotator agreement. Train a U-Net or SegF…
- Semantic Segmentation
- U Net
- Pytorch
Open coursework - CodeIntermediateNew
Distributional Embeddings for a Multilingual Legal Search
Use a public multilingual corpus (e.g., MultiEURLEX or a subset of EUR-Lex) plus a small hand-built test set of around 100 cross-lingual query-passage pairs. Fine-tune (or evalu…
- Distributional Semantics
- Multilingual NLP
- Sentence Embeddings
Computational Semantics - CodeIntermediateNew
Build a LangGraph Multi-Agent Researcher
Design the four-agent topology with explicit message contracts. Implement each agent as a separate LLM call with role-specific system prompts, tool access (web search for resear…
- Multi Agent Orchestration
- Langgraph
- LLM Tool Use
Multi-Agent Systems - 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.
- ResearchIntermediateNew
Hardware-Aware NAS for a Wearable ECG Classifier
You receive a labeled subset of an arrhythmia ECG dataset (about 80,000 10-second windows, 4 classes), a microcontroller latency lookup table (op-level milliseconds) for a Corte…
- Neural Architecture Search
- Hardware Aware Design
- Edge Inference
Open coursework - AnalysisIntermediateNew
Detect Defects on a Production Line for a Tier-1 Auto Supplier
You receive 12,000 labelled grayscale part images (8,000 good, 4,000 defective across 6 defect types) at 2048x2048. Build a pipeline that does classical preprocessing (illuminat…
- Defect Detection
- Image Classification
- Image Preprocessing
Open coursework - CodeIntermediateNew
Train a Reward Model on Customer-Support Preferences
You receive 8,000 labeled preference pairs from real support conversations (each pair is two model responses with a human-chosen winner). Fine-tune a small open-weights base mod…
- Reward Modeling
- Preference Learning
- Bradley Terry Loss
Open coursework - CodeIntermediateNew
Build a Multimodal Generation Pipeline for a Tourism Operator
You receive 40 sample 30-second videos shot by tour guides, the operator's brand voice doc, and SEO keyword lists for EN/PT/ES. Build a pipeline that (1) extracts a representati…
- Multimodal Generation
- Vision Language Models
- LLM Inference
Generative AI 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
- CodeIntermediateNew
Train a GNN for Fraud-Ring Detection at a Payments Fintech
You receive an anonymized transaction dataset (around 120,000 merchants, around 4 million transactions over 12 months, around 2% labeled fraud) and the team's LightGBM baseline.…
- Graph Neural Networks
- Graphsage
- Fraud Detection
Machine Learning on Graphs - CodeIntermediateNew
Extract Skills and Roles from Job Postings for a Recruiter Tool
You receive 30,000 anonymized job postings and a labelled 1,000-posting benchmark with (skill, role, seniority) spans. Fine-tune a small token classifier (e.g., DeBERTa-v3-base)…
- Information Extraction
- Token Classification
- Esco Taxonomy
Open coursework - AnalysisIntermediateNew
Imitation Learning from Human Demos for a Drone Inspection
You receive 6 hours of expert pilot demonstrations (state-action pairs at 20 Hz) recorded in an AirSim wind-farm environment with 3 turbine designs, plus a held-out 4th turbine …
- Imitation Learning
- Behavioral Cloning
- Dagger
Deep Reinforcement Learning - CodeIntermediateNew
Extract Structured Lease Terms for a Commercial Real-Estate Platform
You receive 500 anonymized lease PDFs and a labelled gold set of 150 leases with the 14 fields filled in. Build a pipeline that does (1) layout-aware PDF parsing (Unstructured, …
- Information Extraction
- Pdf Parsing
- Named Entity Recognition
Open coursework - CodeIntermediateNew
Build an Audio-Visual Speaker Diarization Pipeline
Build the pipeline: face detection + active-speaker detection on video, voice-activity detection + speaker embeddings on audio, then a fusion step that ties tracks to detected f…
- Audio Visual Fusion
- Speaker Diarization
- Active Speaker Detection
Multimodal Machine Learning - CodeIntermediateNew
Lambda-Calculus Semantic Parser for a Math-Tutor EdTech
Define a small typed lambda-calculus representation for linear equations and a small set of word-problem templates (rate, age, mixture). Build a parser that maps surface express…
- Semantic Parsing
- Lambda Calculus
- Symbolic Reasoning
Computational Semantics - CodeIntermediateNew
Build a Multilingual Customer-Email Classifier
You receive 28,000 labeled emails (skewed toward English and Mandarin). Try at least two approaches: (1) a fine-tuned multilingual transformer (XLM-RoBERTa or mDeBERTa) and (2) …
- Text Classification
- Multilingual NLP
- Transformers
Open coursework - CodeIntermediateNew
Quantize a CNN for Battery-Powered Wildlife Cameras at a Climate Nonprofit
You receive an FP32 CNN (MobileNetV2 fine-tuned to 22 species, around 13 MB) and a hold-out test set of 4,000 images. Quantize to int8 (post-training quantization first, then qu…
- Quantization
- Qat
- Edge Deployment
Deep Learning - CodeIntermediateNew
Fine-Tune a Transformer for Customer-Support Triage at an Enterprise AI Vendor
You receive 240,000 labeled support tickets across 14 queues, with English, Bahasa Indonesia, and Tagalog. Fine-tune a multilingual transformer encoder (XLM-RoBERTa-base is a st…
- Transformers
- Fine Tuning
- Multilingual NLP
Deep Learning - CodeIntermediateNew
ReAct Agent for Legal-Research Tool-Use
You receive 30 research questions with paralegal-written gold answers and citation lists, plus stubbed implementations of the 4 tools (you do not need to build retrieval — just …
- React Prompting
- Tool Use
- Agent Design
Open coursework - CodeIntermediateNew
Scene-Graph Generation for Retail Shelf Audits
You receive 1,500 labeled shelf photos (anonymized product crops, bounding boxes, and ~12 relation types). Build a pipeline that, for a new shelf photo, outputs (a) detected pro…
- Scene Graph Generation
- Object Detection
- Relation Prediction
Open coursework - ResearchIntermediateNew
Evaluate VAEs vs. Diffusion for Synthetic Tabular-Data Generation
You receive a real labeled dataset (around 18,000 anonymized patient records, 32 features, binary outcome) and the team's existing VAE baseline. Train a tabular diffusion model …
- Tabular Diffusion
- Vae
- Synthetic Data
Generative AI
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.
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.