01 / Overview
Specialist expertise, applied carefully
Resolve Labs works with teams solving hard problems in NLP, machine learning, and software systems. We can assess an approach, build and evaluate a prototype, improve an existing system, or take a focused component through to production.
We are particularly good at applying NLP and ML to language-learning problems.
02 / Capabilities
How we can help
NLP and language technology
Text analysis and generation, writing support, search and classification, LLM applications, model evaluation, and language-learning features.
Machine learning and AI systems
Model and approach selection, data preparation, experimentation, custom pipelines, and integration with existing products.
Evaluation and technical assessment
Feasibility studies, model comparison, error analysis, architecture review, and evidence-based recommendations.
Backend and production engineering
APIs, data pipelines, service architecture, monitoring, and moving prototypes into dependable operation.
03 / Engagements
Ways we can work
Technical assessment
Clarify feasibility, compare approaches, identify risks, and leave your team with an evidence-based recommendation.
Prototype and evaluation
Build a working prototype, define success criteria, and test it against representative data or workflows.
Focused development
Take ownership of a difficult component or feature, from specification through integration and production readiness.
Applied research and experimentation
Investigate a novel or uncertain problem through literature review, experiments, model development, and careful analysis.
04 / Past work
Our team’s work
Selected examples of work delivered by members of our team.
For more about our team members’ backgrounds and experience, see the About page.
ML engineering / Language learning
Interactive translation, one word at a time
As a consultant to PangeaChat, led development of an interactive translation system that let language learners build translations one word at a time from model-suggested options. The system cached language-model decoder states between selections, turned model tokens into usable word choices, and let users efficiently step through decoding one word at a time.
Applied research / Language learning
Top-ranked vocabulary difficulty prediction
Co-developed Sakura, the winning open-track system in the BEA 2026 shared task on vocabulary difficulty prediction. The team fine-tuned LLMs with a novel soft-target loss function for continuous difficulty prediction, then combined three model families with linguistic and cross-lingual features. Sakura ranked first for Spanish, German, and Mandarin learner groups.
Read the paper