We build the tutor behind the exam.
Vulture Consulting LLC is building a personalized 1-to-1 AI tutoring engine for U.S. learners. Each exam gets its own product site. Underneath, every one of them runs on the same engine: a tutor that teaches instead of answering, nudges before a learner drifts, and rewrites the study plan whenever it learns something new.
The engine running inside an example exam product. Content changes per exam; the behavior does not.
The pedagogy is the same for every exam. Only the content changes.
Personal tutoring is the most effective form of instruction we know of, and the least available. Most AI study tools copy the cheapest part of a tutor (answering) and skip the parts that move scores: diagnosing, pacing, following up, and adjusting. We build those parts once, as an engine, and give every exam product the whole tutor.
Learners tutored one-on-one scored about two standard deviations above classroom peers in Benjamin Bloom's studies.
Bloom, "The 2 Sigma Problem," 1984One engine, every exam. Tutoring, nudging, planning, and learner memory are built once and shared by every product we launch.
Platform design goalTo stand up a new exam product, because launching one means loading a blueprint and a question bank, not building a tutor.
Target launch cycleFour steps, repeated every day.
Most AI tutors wait for a question. Ours runs a continuous loop: it understands where a learner is, teaches, follows up, and adjusts.
Diagnose
An adaptive check maps what the learner knows, topic by topic, against the exam's official blueprint and weightings.
Teach
Socratic, step-by-step guidance. Hints and questions first; the full answer only after a real attempt.
Nudge
Check-ins by app, text, or email, timed to when the learner actually studies and tuned to their tone.
Replan
After every session, quiz, or missed day, the plan updates itself and states what changed and why.
After Replan, the loop starts again at Diagnose, using what the tutor just learned.
Watch a study plan adapt.
Pick something that happens in a normal week of exam prep. The planner rearranges the week and the tutor explains the change. This is the same planner every product site ships with.
Example learner: Maya, bar exam in 6 weeks. Choose an event:
Maya's week
Plan v1 · Mon 7:30 amWhat every product built on the engine gets.
Each module is exam-agnostic. A product site turns them on, feeds them its content, and inherits the behavior.
Socratic tutor
Guides with questions and layered hints, grounded in the exam's own materials. Answers come only after an attempt.
tutor
Nudge system
Reminders sent when they are most likely to work, with tone adapted per learner. Quiet hours and opt-outs built in.
nudger
Self-revising planner
Builds a plan backward from the exam date around work, life, and energy. Every change carries a stated reason.
planner
Learner memory
A per-learner skill graph: mastery by topic, recurring mistakes, pace, and how they learn best. Grows across sessions.
memory
Exam blueprint model
The exam's official topic outline, weightings, question formats, and scoring, encoded once and used by every module.
blueprint
Timed practice & projection
Realistic timed sets and full mocks, with a running score projection and the topics most likely to move it.
practice
Snap, speak, or type
Learners can photograph a problem, talk through an answer, or write it out. The tutor works from any of them.
input
Focus & wellbeing check-ins
Short mood and energy check-ins shape the day's load. Signs of serious distress are escalated to a human, never handled by AI alone.
wellbeing
Safety & privacy layer
Content filters, grounding checks against source material, audit logs, and data handling designed for U.S. privacy law.
guard
How a product site is put together.
Four layers. The top one is the only thing that differs between exams.
Exam-specific site
Its own brand, domain, onboarding, and content. Talks to the engine through one API.
↓ engine API ↓
Vulture tutoring engine
Tutor, planner, nudger, memory, practice, and guard modules. Shared by every product.
Exam blueprint & question bank
Official topic outlines, weightings, licensed or original questions, explanations, and source documents for grounding.
Edge-first, model-agnostic
Static front ends on Cloudflare Pages, logic in Workers, data in D1, KV, and R2. Frontier language models with retrieval over the exam corpus; the model layer can be swapped.
Grounded, not improvised
The tutor answers from the exam's own corpus and cites it. Grounding checks flag any explanation it can't trace to a source.
Every decision has a reason
Plan changes, nudge timing, and difficulty choices are logged with the signal that caused them, so learners and reviewers can see why.
Evaluated like a product, not a demo
Each exam ships with an eval set: tutoring quality, grounding accuracy, and plan quality are scored on every model or prompt change.
Model-agnostic by design
The engine talks to language models through one interface. We can move exams to a better or cheaper model without touching product sites.
Privacy first
Learner data stays in the learner's record, is never used for advertising, and is handled under U.S. privacy law, with stricter rules for minors.
Launching a new exam product.
This is the whole playbook. No tutoring logic is rebuilt per exam.
Encode the blueprint
Load the exam's official outline, weightings, formats, and scoring into the blueprint model.
Load the content
Add the question bank, explanations, and source corpus. Run the grounding and eval suites.
Brand the shell
Stand up the exam's own site and domain on the shared front-end template.
Launch and learn
Ship, watch the evals and learner outcomes, and feed improvements back into the engine for every product.
Based in Florida. Built for U.S. learners.
Vulture Consulting LLC is a U.S. technology company headquartered in Florida. We build and operate the engine and the exam products that run on it.
- What we are
- An engineering and learning-science company. Our product is the engine; exam sites are how it reaches learners.
- Focus
- U.S. high-stakes exams first: professional licensure, graduate admissions, and certifications, where learners need a plan and a coach, not just a question bank.
- Compliance
- U.S. privacy law from day one: state privacy statutes, FERPA where school data is involved, and COPPA-grade handling for any learner under 13.
- Team
- Florida-based, remote-friendly. Engineering, learning science, and exam content specialists.
Engine first, then breadth.
Engine v1
Tutor, planner, nudger, and memory modules. Blueprint model and eval harness.
First exam products
Launch the first exam sites on the engine. Measure learner outcomes and tutoring quality.
Multi-exam platform
Shared front-end template, content tooling, and a launch playbook that takes weeks, not months.
Partner access
Open the engine API to course providers and institutions that want the tutor inside their own products.
Building an exam product? Studying the problem? Talk to us.
We work with exam content experts, course providers, institutions, and researchers. We're also happy to walk investors and partners through the engine in detail.
- Partnerships and engine access
- Exam content and learning-science collaboration
- Press and research inquiries