Project Overview
Educational institutions that want rigorous, course-specific exams usually have to write every question by hand, and students preparing for those exams rarely get to practice under conditions that match the real thing. ExamSim, a Villaex Tech product, addresses both sides of that gap: institutions use it to generate accurate, tailored multiple-choice questions with AI, and students use it to practice under timed, realistic conditions with instant feedback.
At a Glance
- ▹Business goal: Give educational institutions a faster way to produce accurate, course-specific exam questions, and give students a realistic space to practice for them
- ▹Primary users: Institutions generating assessments for their own courses, and students preparing for real exams
- ▹Core solution: A multi-tenant SaaS platform where institutions generate AI-powered MCQs and students practice them in a timed, instantly-graded simulator
- ▹Key outcome: A live product in use by real institutions and their students, with reliable AI question generation as its core differentiator
Client or Business Context
ExamSim is Villaex Tech's own SaaS product, built for educational companies that need to produce exam content at scale. It's structured as a B2B platform: institutions purchase access under a tiered model and manage their own student cohorts independently, without their content or users overlapping with other institutions on the platform.
Project Positioning
This is a startup product — developed by the Villaex Tech team as a commercial offering for the education sector, rather than a one-off client engagement. I was part of the team that built it, focused specifically on the AI question generation and validation pipeline.
The Challenge or Opportunity
What Was Not Working
- ▹Producing high-quality, course-specific exam material by hand is slow and expensive for educational companies
- ▹Students preparing for exams often lack a practice environment that reflects the real exam's format, pacing, and difficulty
Why It Mattered
When every question set has to be written manually, institutions face a real bottleneck on cost and speed every time they need new or updated assessments. And without a realistic environment to rehearse in, students go into real exams without an accurate sense of what they're actually walking into.
Project Goals
- ▹Let institutions generate large volumes of accurate, appropriately difficult MCQs for any course, on demand
- ▹Give students a realistic, timed practice environment with instant feedback
- ▹Support multiple institutional clients on one platform, each managing their own cohorts independently
- ▹Make AI-generated content reliable enough to trust for real, enterprise-grade assessment use
The Approach
1. Understanding the Problem
The core insight behind ExamSim was that the exam-content bottleneck and the lack of realistic student practice were really two sides of the same problem: institutions needed a faster way to produce trustworthy questions, and students needed somewhere realistic to use them. The platform was scoped to solve both rather than just one.
2. Defining the Right Scope
Key scope decisions:
- ▹Multiple-choice questions as the initial format: keeping generation scoped to MCQs, rather than open-ended formats, made grading deterministic and made it possible to validate AI output automatically before a student ever saw it
- ▹Multi-tenant from the start: the platform was built with tiered purchasing and access control from day one, so it could support more than one institution without needing to be re-architected later
3. Designing the Solution
Core solution components:
- ▹AI Question Generation: institutions generate hundreds of tailored multiple-choice questions from specific course parameters, instead of writing them by hand
- ▹Realistic Exam Simulation: a distraction-free testing environment for students, with timers and immediate grading
- ▹B2B SaaS Architecture: tiered purchasing and access control that let institutions manage their own student cohorts independently on shared infrastructure
4. Building and Validating
My focus was the layer institutions rely on to trust the platform at all: the AI question generation pipeline and the backend validation behind it. That meant structuring how prompts translated course parameters into questions, and building checks that ran against every generated question before it reached an institution.
5. Solving the Hardest Problem
Challenge: Making AI-generated questions reliable enough for enterprise use
Why it was difficult: Large language models can produce questions that are subtly incorrect, ambiguously worded, or inconsistent in difficulty — none of which are acceptable when an institution is using the output for a real assessment.
Decision made: I engineered the prompt architecture behind question generation and built backend validation on top of it, checking every AI-generated question for factual accuracy, correct formatting, and consistent difficulty before it was surfaced to an institution.
Outcome: Institutions can generate hundreds of tailored MCQs per course and use the output directly, rather than needing to manually review and rewrite AI-generated questions before they're usable.
The Solution
ExamSim gives institutions a way to generate exam content on demand and gives students a realistic space to practice with it. Institutions set course parameters and receive AI-generated MCQ sets they can trust; students then work through those questions in a timed, distraction-free simulator with instant grading.
Key Capabilities
- ▹AI Question Generation: instantly generates hundreds of tailored multiple-choice questions based on specific course parameters
- ▹Realistic Exam Simulation: a distraction-free testing environment for students, featuring timers and immediate grading
- ▹B2B SaaS Architecture: tiered purchasing and access control, letting institutions manage their own student cohorts independently
Key Results
ExamSim is live and in use by real institutions and their students. Formal usage metrics aren't tracked yet, so the results below reflect what's been delivered and validated rather than measured numbers.
Product and User Outcomes
- ▹Live in production, used by real educational institutions and their students rather than sitting in a demo or pilot state
- ▹Institutions can generate large volumes of tailored MCQs for their own courses on demand, replacing manual question writing for at least part of their assessment workflow
- ▹Students get a distraction-free, timed simulation of real exam conditions with instant feedback, rather than practicing in a format disconnected from the actual exam
Technical Outcomes
- ▹Output reliability: built validation logic that checks every AI-generated question for factual accuracy, correct formatting, and appropriate difficulty before it reaches an institution
- ▹Scalable generation: structured the question-generation pipeline to reliably produce large volumes of usable questions, rather than requiring manual review of each output
- ▹Multi-tenant support: the platform serves multiple institutions independently, each with their own cohorts and access tier
Qualitative Results
- ▹Now used in production by real institutions and their students
- ▹Replaced manual, one-by-one question writing with on-demand AI generation for course-specific assessments
- ▹Gave students a practice environment that mirrors real exam conditions instead of a generic quiz format
Impact
For institutions, this means less time spent authoring questions by hand for every course, with enough confidence in the AI-generated output — thanks to the validation layer behind it — to use it for real assessments rather than treating it as a rough first draft. For students, it means exam practice that actually resembles the real thing: timed, graded instantly, and specific to their course, instead of a generic study tool disconnected from the actual test format. Because the platform is multi-tenant, this holds for more than one institution without needing a separate build per client.
Skills Demonstrated
- ▹Integrating large language models into a live, commercial product pipeline
- ▹Prompt engineering for structured, dependable AI outputs
- ▹Building validation logic to catch inaccurate, malformed, or inconsistent AI-generated content before it reaches users
- ▹Backend data pipeline design supporting a multi-tenant SaaS product
My Role and Contribution
I was part of the team that built ExamSim at Villaex Tech, focused specifically on the AI question generation system and the backend validation pipeline behind it.
I was responsible for:
- ▹Designing the prompt architecture used to generate MCQs from course-specific parameters
- ▹Building the backend validation layer that checks AI-generated questions for factual accuracy, correct formatting, and appropriate difficulty before they reach an institution
- ▹Structuring the AI generation pipeline to reliably produce large volumes of usable questions, rather than requiring manual review of every output
Team setup: Part of the Villaex Tech team. I owned the AI/LLM question generation and backend validation pipeline; other team members built the student-facing exam simulation experience and the multi-tenant access control and purchasing system.
Lessons and Reflection
- ▹What worked well: Building validation directly into the generation pipeline, rather than treating it as a separate review step afterward, meant bad output could be caught automatically before an institution ever saw it.
- ▹What I would improve: We don't track formal usage metrics yet. I'd push to add that instrumentation so we can see exactly how the generation and validation pipeline performs across institutions, rather than only knowing anecdotally that it's working.
- ▹What the project taught me: Working on the reliability layer for AI-generated content taught me that trustworthy AI output in a commercial context depends far more on the validation and structure built around the model than on the model itself.
Closing Summary
ExamSim gives educational institutions a way to generate accurate, course-specific exam questions on demand, and gives students a realistic space to practice under real exam conditions. Built by the Villaex Tech team — with the AI question generation and backend validation pipeline as my primary contribution — the platform is live today and in use by real institutions and their students.
Project takeaway: Made AI-generated assessment content reliable enough for institutions to trust in a real, commercial product.
