BTLPT Online Course & AI Practice System
A complete preparation ecosystem for the Bilingual Target Language Proficiency Test (BTLPT), serving both individual candidates and university certification programs across Texas. The system combines a preparation book, an AI-powered online course, a full-length practice exam, and a custom AI evaluation engine into a single integrated platform.
The core design principle is adaptive, criterion-referenced evaluation grounded in Vygotsky's Zone of Proximal Development. The system meets each learner where they are, adjusting difficulty and feedback based on demonstrated performance, so practice is always challenging but achievable.
I designed and built every component: the curriculum, the web applications, the AI evaluation pipeline, the professor dashboards, and the institutional delivery model. This is not a chatbot wrapper. It is a code-driven system built on multiple large language models (Claude Sonnet, GPT-4o) with retrieval augmented generation, embeddings, and structured rubric schemas custom-built for this exam.
Impact Numbers
Books sold on Amazon (4.6-star rating, 2nd edition)
Online learners enrolled across individual and institutional accounts
Full-length practice exams completed with instant AI feedback
Universities and school districts using the system
Institutional Partnerships
Universities and certification programs across Texas
Dallas College, East Texas A&M University, Texas A&M International University, Texas A&M University-San Antonio, Texas Tech University, Texas Woman's University, University of Houston-Clear Lake, University of North Texas at Dallas, Dallas ISD, Garland ISD, McAllen ISD, and others.
System Architecture
Four interconnected components form a complete preparation pathway, from foundational study to exam-day simulation.
BTLPT Preparation Book
The #1 Amazon resource for BTLPT preparation, now in its 2nd edition. Covers strategies for all four exam sections (listening, speaking, reading, writing), academic vocabulary, grammar rules, and model responses. Includes a digital extension with online videos, practice activities, and AI-guided study.
AI-Powered Online Course
12 instructional modules with 137 tasks covering all four BTLPT domains. AI-powered oral practice (record, listen, submit for evaluation), writing practice across three task categories (correspondence, lesson plans, essays), and accent mastery with 200+ words and adaptive difficulty. Every practice session generates a new, unique challenge.
Full-Length Practice Exam
A timed simulation of the entire BTLPT across all four sections. Three exam versions available. Individual candidates receive instant AI scoring; university programs get a professor evaluation panel where faculty can listen to oral recordings, review written responses, and override AI scores with personalized comments. Timed access codes simulate real testing conditions.
Veronica: AI Evaluation Engine
A custom-built AI evaluation system (not a chatbot) that powers every practice activity. Built on multiple large language models (Claude Sonnet, GPT-4o) with retrieval augmented generation, embeddings, and structured rubric schemas tailored to the BTLPT. Scores oral responses on the 0-9 scale and written responses on the 100-300 scale across three criteria: task fulfillment, organization, and language use. Detects code-switching and non-standard forms, provides quoted corrections, and generates model answers for comparison.
What I Built
Technical highlights across the system's design, development, and deployment.
AI Evaluation Pipeline
- ✓ Multi-model LLM architecture with RAG and embeddings
- ✓ Structured rubric schemas custom-built for the BTLPT
- ✓ Code-switching and non-standard form detection
- ✓ Adaptive difficulty based on learner performance
- ✓ Quoted error identification, corrections, and model responses
Institutional Features
- ✓ Faculty dashboards to monitor student progress, scores, and completion across cohorts
- ✓ Professor grading panel: listen to oral recordings, review writing, comment, grade, and override AI evaluations
- ✓ Timed access codes for authentic exam simulation
- ✓ Cohort management, CPE certificates, and digital credentials (POK API integration)
Technical Stack
- ✓ Multiple LLMs (Claude Sonnet, GPT-4o, Whisper) with RAG and embeddings
- ✓ WordPress + LearnDash LMS with custom PHP mu-plugins
- ✓ PHP and JavaScript single page app architecture
- ✓ Web Audio API, REST endpoints, WooCommerce
- ✓ POK digital credentials API integration, AWS SES email automation
Key Differentiators
What separates this from typical test prep platforms.
Adaptive learning grounded in ZPD
The system adapts to each learner's performance based on Vygotsky's Zone of Proximal Development. Difficulty, feedback depth, and practice focus adjust automatically so students are always working at the edge of their ability, not repeating what they already know.
Custom AI architecture, not a chatbot
Multiple large language models (Claude Sonnet, GPT-4o) orchestrated through custom PHP and JavaScript code with RAG, embeddings, and structured rubric schemas designed specifically for the BTLPT. Multi-model cross-verification ensures consistent, criterion-referenced evaluation across every response.
Fresh content every session
AI generates new, unique practice prompts for every attempt through structured output schemas that ensure each response meets the exam's domain specifications. Students never see the same scenario twice, preventing memorization and ensuring authentic practice.
Dual evaluation: AI + Professor
AI provides instant rubric-based feedback. Professors can then listen to oral recordings, review written responses, comment, grade, and override AI evaluations with their own scores and personalized guidance. Students benefit from both automated and human evaluation.
Built for institutions, not just individuals
Cohort management, faculty dashboards with student progress monitoring, timed access codes, CPE certificates, and verified digital credentials. Currently partnering with 16+ universities and school districts across Texas.
Linguistically specialized
Grounded in academic Spanish conventions, TEKS, ELPS, and bilingual pedagogy standards. Detects code-switching, non-standard forms, and orthographic patterns specific to heritage speakers of Spanish.
End-to-end ownership
I authored the book, designed the curriculum, built the web applications, developed the AI evaluation pipeline, and manage the institutional partnerships. Full-stack instructional design and development.
Explore the System
The BTLPT preparation system is live and actively used by educators across Texas. Visit the course page or contact me for institutional partnership details.