The math tutor that actually teaches.
An AI tutor for Texas middle school math — TEKS-aligned, designed for HB 1416 accelerated instruction, and architected to refuse to give students direct answers. Because answers don't teach. Questions do.
Prototype — early accessMost AI for schools is a horizontal chatbot with a FERPA badge. We're different. MillionRoots is an AI math tutor built for one subject, one grade band, one state — middle school math, aligned to Texas TEKS, designed to meet HB 1416 accelerated instruction requirements. The tutor uses hint-ladder pedagogy, refuses to give direct answers, and gives teachers a real-time view of where every student is stuck.
Built for the Texas classroom you actually teach in.
TEKS-native, not TEKS-adapted
Most "standards-aligned" math products are Common Core products with state-specific mappings bolted on. MillionRoots is the opposite — TEKS is the spine of the product. The tutor knows 6.4D means ratio reasoning and 7.4A means proportional relationships. Your pacing guide and your vendor finally speak the same language.
Designed for HB 1416 accelerated instruction
HB 1416 requires 15–30 hours of accelerated instruction for students who don't meet STAAR benchmarks. Most districts don't have the teachers to deliver that. MillionRoots is designed as scalable supplemental instruction that meets the intent of the law — one-on-one attention, standards-aligned, teacher-reviewable — without asking you to hire people who aren't there to hire.
SB 2124 advanced mathematics pathway
SB 2124's advanced math program requires districts to offer a pathway to Algebra I by 8th grade. MillionRoots supports that pathway — accelerated problem sets, prerequisite-gap diagnostics, and mastery tracking mapped to the advanced TEKS sequence. Your most-prepared students get more challenge; your emerging students get the scaffolding that makes acceleration honest rather than cosmetic.
See what actually happens when an AI refuses to give answers.
Type a question below. The tutor will respond in real time using the same hint-ladder pedagogy it uses in the full product. Try asking for the answer directly. Try saying "I don't know." Try being the toughest student you can imagine. The tutor won't break.
- Runs on real AI, no scripted responses
- Adversarial inputs are fair game — the pedagogy survives pressure
- 5 exchanges here — open the full demo for more
The hint ladder, shown on one problem.
A real exchange, abstracted. Same tutor, same refusal-to-gift-answers behavior you just tried above.
- Step 1Student What's the answer?Tutor I could tell you, but you'll learn more if we figure it out together. What does the ratio 3:5 mean to you?
- Step 2Student 3 boys and 5 girls?Tutor That's one way to read it. Now, the class has 32 students — how many 'groups' of 8 would fit into that?
- Step 3Student 4 groups.Tutor Exactly. And each group has 3 boys. So how many boys are in the class?
This is how real tutors teach. Not by giving answers. By asking the next right question. We call it the hint ladder, and we've encoded it into the AI — refuses to gift answers, escalates support when students genuinely stall, and adapts to where each student actually is.
Every conversation is visible. You stay the teacher.
Every tutor conversation is reviewable. Every student's stuck points are flagged in real time. Every AI-suggested intervention is yours to approve or reject before it reaches a student. You see what the tutor saw. You see what the student tried. And you decide what happens next.
- Real-time view of who's working, who's stuck, who's done
- AI-detected pattern flags across your whole class
- One-click access to any student's full conversation history
- Nothing happens to a student's instruction without your review
Private by architecture, not by policy.
- No student data ever trains any AI model, ours or anyone else's
- FERPA-aligned, COPPA-aligned, SOC 2 roadmap in progress
- Teacher-reviewable audit logs for every student conversation
- District-scoped data, no cross-district pooling, ever
- Texas data residency on request
- Founder-accessible escalation path for data concerns
The Privacy & Procurement Checklist
25 questions across five areas — data handling, compliance posture, architecture, classroom safety, operational readiness. Use it as your evaluation framework when reviewing any AI vendor, not just us.
Download PDF (no email required)4 pages. No form, no gate — download and go.
From the founder.
I'm Vijay. I'm building MillionRoots because when I asked a middle school math teacher what AI was doing in her classroom, she told me it was making her kids weaker, not stronger. I wanted to see if we could build the opposite.
This is a solo, founder-led company right now — and that's a choice, not a limitation. Independent means no investor pressure to pivot to breadth when we should go deeper. No quarterly targets dictating what your students see. No acquisition exit that leaves you migrating to a different vendor in two years. I'm building this for the long run, and if you pilot with MillionRoots, you're working with someone who'll still be answering your emails in 2029.
If you're a Texas teacher, principal, or parent and something here resonates, I'd love to hear from you.
Considering a pilot?
We're working with a small cohort of Texas schools and tutoring centers in fall 2026. Pilots are free. You keep all student data. You shape the product. We're looking for math departments who want to be design partners, not just customers.