browse paths latest start your path→

loading path…

curated, practical paths through ai/ml. no noise, no committees.

home generate ogbuilds ↗built in public

← choose a different focus

nlpmixed~4 weeks, part-time

nlp & the transformer

the modern nlp stack from the ground up: represent text, understand the transformer deeply, then fine-tune and ship.

4 modules · 13 curated resources · checkpoint per module

mastery0/4 modules

pass each module's checkpoint to master it.

explore more nlp paths, or generate one for a different topic.

generate a path→

stay current

what's new in nlp

see the full digest →

CADI: A Cross-source Alignment and Dynamic Knowledge Injection Framework for Medical QAintroduces a framework that aligns knowledge from multiple medical sources and dynamically injects it into qa systems to improve accuracy. for practitioners building medical qa tools, this addresses the challenge of integrating diverse, authoritative knowledge sources without manual curation.
Can small language models handle context-summarized multi-turn customer-service QA? A synthetic data-driven comparative evaluationevaluates whether smaller language models can handle multi-turn customer service conversations when context is summarized, using synthetic data for benchmarking. this is practical for teams wanting to deploy qa systems at lower cost and latency without relying on large models.
UniFuse: Unified prototype refinement and calibration for few-shot named entity recognitionproposes a unified framework for refining prototypes and calibrating confidence in few-shot ner tasks, where you have very few labeled examples per entity type. this matters because many real-world ner deployments face data scarcity, and better calibration helps you know when the model is actually uncertain rather than overconfident.