The situation
The edtech (~45 staff) operates across Spain, Portugal, and Italy and uses career-path insights as a differentiator with university partners.
Your task
Turn raw job-posting text into ESCO-normalized (role, skill, seniority) tuples with measurable end-to-end accuracy.
The brief
You start with
You receive 30,000 anonymized job postings and a labeled 1,000-posting benchmark with (skill, role, seniority) spans.
What you do
- Fine-tune a small token classifier (e.g., DeBERTa-v3-base) for span extraction.
- Build a normalization layer that maps extracted skill strings to ESCO skill URIs (fuzzy match + embedding similarity).
- Evaluate per-class F1 and end-to-end (extraction + normalization) accuracy on the held-out benchmark.
You deliver
: pipeline code, fine-tuned model, ESCO mapping notebook, and a 3-page deployment recommendation including a refresh strategy when ESCO updates.
What you will practice
- Fine-tune a token classifier on a real IE task
- Normalise extracted entities to a public taxonomy (ESCO)
- Evaluate end-to-end IE + normalization accuracy
- Design a refresh strategy for taxonomy-dependent pipelines
A suggested way in
- Use Hugging Face TokenClassification trainer; DeBERTa-v3-base is a strong baseline
- Pre-compute ESCO skill embeddings once and cache them
- Report end-to-end accuracy — span extraction can be perfect and normalisation can still drop accuracy
Skills you will practice
Information extraction, Token classification, Esco taxonomy, Entity normalization, Fine tuning, Evaluation
Where it leads
NLP Engineer, Machine Learning Engineer, Data Engineer
How it is assessed
- Extraction quality25%
Per-class span F1 above target on held-out set.
- Normalisation quality25%
ESCO mapping accuracy reported with confidence intervals.
- End-to-end evaluation25%
End-to-end accuracy is reported alongside stage-wise metrics.
- Deployment plan25%
Refresh strategy for ESCO updates is concrete and operable.
Resources
- PyTorchTool
Train the token-classification head.
- Hugging Face TransformersTool
Token-classification trainer and model loading.
- sentence-transformersTool
Embed ESCO skill descriptions for similarity matching.
- PythonTool
Glue the IE and normalisation stages.
- Weights & BiasesTool
Track training and eval runs.
- NotionTool
Author the deployment recommendation.
Finish it. Earn this.
A verified certificate with your name. Anyone can check it.
Time to practice.
Solve this challengeMore like this.
- Software EngineerBuild a Multi-Level Feedback Queue Scheduler in xv6
- Software EngineerReverse-Engineer a Stripped Binary to Restore Linking Symbols
- Software EngineerImplement a Distance-Vector Routing Algorithm
- AI EngineerEnd-to-End Lane Following on a Donkeycar Platform
-
Build a Multi-Level Feedback Queue Scheduler in xv6 Software Engineer Fork the xv6-riscv repository and implement MLFQ in proc.c with 3 priority queues, per-queue time slices (10ms / 20ms / 40ms), priority demotion on sl operating systemsschedulingc programming Open challenge → -
Reverse-Engineer a Stripped Binary to Restore Linking Symbols Software Engineer Receive the stripped shared object (Linux ELF, x86-64, ~280KB), a crash dump with 12 frames of indecipherable addresses, and 4 years of partial git hi reverse engineeringlinking and loadingmachine code Open challenge → -
Implement a Distance-Vector Routing Algorithm Software Engineer Implement distance-vector routing in Python: each simulated router maintains a distance vector, exchanges vectors with neighbors every 1 second of sim routingdistance vectornetwork simulation Open challenge → -
End-to-End Lane Following on a Donkeycar Platform AI Engineer Use the public Donkeycar Tub dataset (or collect about 30 minutes of driving on the simulator). Train a CNN-policy baseline (the Donkeycar default arc end to end learningimitation learningpytorch Open challenge →