# Va Nlp Engineer

> Use when building production NLP systems, implementing text processing pipelines, developing language models, or solving domain-specific NLP tasks like named entity recognition, sentiment analysis, or machine translation.

- Skill: `roche-k/va-nlp-engineer` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx skillmds@latest add roche-k/va-nlp-engineer`
- Raw SKILL.md: https://api.skillmd.com/api/skills/roche-k/va-nlp-engineer/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: roche-k (https://skillmd.com/u/roche-k)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/roche-k/va-nlp-engineer

---


# Nlp Engineer

Use this skill for the capability described above and keep the user's requested
scope, logic placement, and existing project conventions. Read the relevant
sections of [the domain guide](references/guide.md) for detailed considerations;
select the checks that affect this task instead of treating the guide as a
mandatory full-project checklist.

## Workflow

1. Identify the actual data sources, schemas, ownership, processing stages, and acceptance criteria.
2. Work through the existing pipeline or model interfaces, accounting for missing data, leakage, and reproducibility.
3. Evaluate against observed data and an appropriate baseline; report uncertainty and limitations.

## Task-specific focus

- Review existing text processing pipelines and model performance
- Analyze language requirements, domain specifics, and scale needs
- Implement solutions optimizing for accuracy, speed, and multilingual support

## Applying the guidance

Use only tools and services actually available in the session. This skill is
instructional guidance; it does not create a subagent, grant tool permissions,
select a model, or supply a context-manager service. Related specialists are
optional; this skill works independently.

Treat source performance numbers, coverage percentages, and version examples as
context, not verified results or universal acceptance gates. Preserve the user's
test and artifact rules. Report only observed outcomes and disclose checks that
could not run. Produce the requested deliverable without creating extra planning
or status files unless the user requests them.