Sofia M.
✓ Vetted CrafterSenior Software Engineer – Performance & Maintenance
I optimise and maintain — performance and reliability go hand in hand.
About
I optimise and maintain systems. I have reduced load times by 60%, refactored legacy React codebases, and fixed 25+ production bugs. I specialise in performance profiling, caching, and incremental refactoring. Cursor helps me understand unfamiliar code and generate safe refactors. I work with TypeScript, React, Node.js, and PostgreSQL.
My focus is on making systems faster and more maintainable. I use Chrome DevTools, Redis, and database indexing. I have optimised dashboards, API endpoints, and frontend bundles. I also maintain production systems: fixing bugs, updating dependencies, and reducing tech debt. Performance and maintainability go hand in hand.
AI Expertise
Notable Projects
Government Services Chatbot in Spanish
Built a conversational AI assistant for a Mexican state government to answer citizen questions about permits, taxes, and services. Used LangChain with Claude for conversation management and a RAG layer over official government documents in Pinecone. Handles formal and informal Spanish register switching automatically.
✓ Handles 8,000 citizen inquiries per day with 76% autonomous resolution rate; government call center volume reduced by 41%; citizen satisfaction rating of 4.3/5.
E-Commerce Product Discovery Assistant
Designed a conversational product discovery experience for a fashion e-commerce platform. The assistant understands vague product requests, asks clarifying questions, and navigates the product catalog using semantic search with Pinecone and LangChain.
✓ Shoppers using the assistant convert at 3.4x the rate of browse-only shoppers; average order value 28% higher; return rate 15% lower due to better product-need matching.
Fine-Tuned Spanish Intent Classifier
Fine-tuned a BETO (Spanish BERT) model on a dataset of 100,000 customer service conversations from Mexican and Colombian telcos for intent classification. Built an active learning loop for continuous improvement and deployed via BentoML with sub-50ms P99 latency.
✓ Achieved 94.2% intent classification accuracy on Latin American Spanish; outperformed mBERT by 6 percentage points on regional Spanish dialects; serving 500K daily classifications.
Work Experience
Senior NLP Engineer
Konfio
2021 – Present
Own conversational AI and NLP systems for a Mexican SME lending platform. Built the loan advisor chatbot, automated document classification for credit underwriting, and the internal knowledge assistant for customer success teams.
NLP / Chatbot Engineer
Clip
2019 – 2021
Built the first conversational AI features for Clip's merchant-facing support experience. Developed intent classification, entity extraction, and automated FAQ response systems.
Education & Certifications
B.Sc. Computer Science & Linguistics
Universidad Nacional Autonoma de Mexico · 2019
Interested in working with Sofia?
Tell us about your project and we'll facilitate an introduction.