Industries are full of repetitive work people don't need to be doing.
We build LLM and multimodal AI products for industries drowning in repetitive cognitive work. We're starting with Indian education — where teachers spend 30–40% of their time evaluating descriptive answers, often handwritten, often in regional languages.
Our story
The name comes from two words: Arivu, Tamil for intelligence or wisdom, and Yantra, Sanskrit for machine. Ariantra — an intelligent machine.
We're both mechanical engineers by training. “Foundry” is a nod to where we started — shaping raw material into something useful. After 15 years in software and operations, we're doing the same thing now, just with a different material.
Ariantra AI was started by two operators with 15+ years of experience each in product and operations roles — long enough to see, up close, how much time gets lost to repetitive work that could be automated but isn't. The question we kept returning to: where can technology actually give back time to the people who create value, instead of adding another tool to manage?
We started looking at education because that's where the answer was loudest. In conversations with educationists across India, the same pain surfaced again and again — 30 to 40 percent of a teacher's time goes into evaluating descriptive answers. Assignments are quietly outsourced to parents at home. The dedicated “evaluation teachers” who once shouldered this load are vanishing. Teaching is being eaten by correction work.
Education first. Other heavy-cognitive-work industries over time.
- Arivu
- Tamil — intelligence, wisdom
- Yantra
- Sanskrit — machine
What we build
An AI product is only as good as the engineering behind it. Here's what powers our products today.
Multi-stage AI pipeline for descriptive answer evaluation
LLM-based scoring against rubric criteria, with anti-hallucination verification and confidence scoring on every grade.
Multimodal handwriting recognition
OCR plus vision-LLM escalation for handwritten answer sheets, including Hindi, Tamil, Telugu, and Kannada scripts alongside English.
AI orchestration for assessment workflows
Question-paper generation across 17 question types (Bloom's-taxonomy aware), proctored online assessments, and teacher-review workflows.
Multi-tenant SaaS architecture
Built on MongoDB with strict per-tenant data isolation, designed to scale across thousands of schools and exam-season load spikes.
Cloud-native infrastructure
Object storage, async job processing, connection-pooled databases, designed for elastic scale.
Built for Indian education
CBSE, ICSE, and state boards. Regional-script handwriting recognition is part of the product today, not a roadmap line.
R&D roadmap
Our roadmap includes self-hosted Indic-language handwriting OCR, fine-tuned domain models for regional-language descriptive evaluation, and on-premise deployment options for boards and government — all of which require dedicated AI infrastructure. We're investing here because regional-language evaluation at the depth Indian classrooms need cannot be served by general-purpose foundation models alone.
MarksZen
Built specifically for Indian boards — CBSE, ICSE, and state — with regional-script handwriting recognition as a first-class capability.
MarksZen is our AI grading platform. It evaluates descriptive answers (including handwritten ones), generates question papers in seconds, and runs proctored online assessments.
Read more about MarksZen →Founders
Where we are
We're shipping a live product to schools and individual teachers, building the AI infrastructure for Indian-language evaluation, and open to conversations with mission-aligned partners.
Company details
- Registered name
- Ariantra AI Foundry Private Limited
- Registered office
- 400-A, 4th Floor, Yusuf Sarai Commercial Complex, Hauz Khas, New Delhi – 110016
- CIN
- U63119DL2026PTC465754
- PAN
- ABFCA4809E
- TAN
- DELA88582E
- Date of incorporation
- 21 April 2026