Enhance Spanish AI Accuracy

Nexoply.com serves a growing Spanish–speaking audience, yet the current AI module drifts in accuracy whenever conversations switch from English to Spanish. I want the model’s Spanish behaviour brought to the same standard as its English output, with special attention on the relevance and factual soundness of every reply.

Primary focus
• AI response relevance, specifically the accuracy of each generated answer.

Current pain-points being observed
• Grammar and spelling errors
• Context misunderstanding
• Incorrect information

Scope of work
1. Audit the existing NLP pipeline and training data that power the Spanish responses.
2. Pinpoint failure cases causing the three issues above, then retrain, fine-tune or augment the model accordingly (e.g., domain-specific datasets, prompt engineering, reinforcement-learning-from-feedback).
3. Refactor or extend the backend code so the improvements deploy cleanly into our serverless Node environment.
4. Validate the upgraded model with side-by-side A/B tests in production traffic until the Spanish accuracy rate meets or exceeds the English baseline.

Acceptance criteria
• ≥ 95 % reduction in the listed inaccuracies across a 1,000-prompt evaluation set.
• All changes fully documented and version-controlled so future iterations can replicate your workflow.
• A concise report summarising methodology, datasets used, metrics before/after, and next-step recommendations.

Tech you should feel comfortable with: Python, Node.js, OpenAI / Hugging Face APIs, vector databases, and CI/CD on Vercel. Access to our private GitHub repo and staging keys will be provided after project kickoff.

Turnaround is flexible so long as we see measurable progress each week.