Menu

Embodied AI, built in India

Robots that hold a room.

PredictML builds embodied-AI machines that work in the places people actually are — classrooms, reception halls, hospital corridors. Two flagship robots, one edge-AI platform.

Request a demoExplore EDUBOT

Footage: EDU-BOT taking a lesson at Mandal Parishad High School, Kokapet. Fixed upper body, wheeled base, no articulated limbs.

One platform, two flagships

The same machine underneath.

EDUBOT and AURRA BOT are two expressions of one hardware architecture. Select a component to see what it does.

EDUBOT seen full height: a tapered white body with gold lettering, a chest-mounted touchscreen and a ring-lit camera head
Platform components

Jetson Orin — Runs perception, speech and navigation on the machine itself, so a classroom with unreliable connectivity still gets a working robot.

LiDAR — Maps the room continuously and keeps the robot a safe distance from people and furniture.

RGB camera — Recognition and scene understanding, feeding the interaction layer rather than a remote server.

Touchscreen — Set at the eye level of a seated student or a person at a counter — the primary way people interact.

BLDC drive — Moves the robot between positions. The upper body is fixed; there are no articulated limbs.

ESP32 — Handles the real-time control loop underneath the AI stack.

Two robots, two rooms.

Schools

EDUBOT

A teaching assistant for K–12 classrooms. Delivers interactive lessons, runs quizzes, takes attendance and answers questions in English and regional languages.

Explore EDUBOT

Hospitals and front offices

AURRA BOT

A front-of-house robot for reception halls and hospital corridors. Greets visitors, answers routine questions and guides people to where they need to be.

Explore AURRA BOT

Also from PredictML

Two platforms, running in software.

The same engineering, without the chassis: an inference gateway that keeps routine AI work on local GPUs, and a marketing team of AI agents that ships nothing without your approval.

Hybrid multi-modal AI gateway

prouter

prouter.io

The Local-First Intelligent AI Gateway.

An autonomous L7 inference gateway that predicts prompt workload complexity in real time, routing routine tasks to local GPU clusters at $0 per token and escalating complex reasoning to cloud APIs under strict budget governance.

Modelled cloud savings
86%
Route overhead goal
<15ms
Modalities: text, image, video, voice
4
  • Unified APIOne OpenAI-compatible gateway for text, image, video and voice workloads.
  • Predictive workload routingA host-level classifier evaluates prompt intent, token length and structural complexity before execution, so low-complexity tasks stay on local GPUs.
  • Dynamic cascadingComplexity scoring keeps routine traffic local and escalates only when the request needs it.
  • Bare-metal arbitrageSaturates RTX 3090 and A6000 capacity before spending wallet credits on cloud APIs.
  • Stateful KV-affinity cachingHashes system prompts, persistent files and RAG contexts so follow-up requests stick to the worker node already holding the cache, eliminating prefill overhead.
  • Governed cloud escalationPer-user daily budgets, approval workflows and token caps; cloud calls execute only when local capacity or capability thresholds are exceeded.
Launch prouter (opens prouter.io in a new tab)

Marketing agent platform

reAIagents

reaiagents.com

Six AI teammates run your marketing. You approve what matters.

Six specialist AI agents plan, create, run and optimise marketing end to end — one coordinated team on a single shared model of your brand, where every send and every dollar waits on your approval.

  • Approval before anything shipsEvery agent's ability to publish, send or spend is off by default. An agent proposes, you approve — or you don't, and nothing happens.
  • Six specialists, one systemElara leads and delegates; Sora content and social; Kael growth and advertising; Lyra research and SEO; Orin leads and outreach; Mira analytics and optimisation.
  • Brand DNA, from almost anythingPoint it at a website, or upload a brand deck, a logo or a pitch video — it extracts voice, palette and positioning either way.
  • A library that reuses before it regeneratesNew creative is only generated when nothing existing already fits: search, then generate, not generate-by-default.
  • Lifecycle sequences on your own contactsDrip sequences and churn triggers run on the customers you already have, not a purchased list.
  • A real marketing skills librarySEO, growth, retention and sales playbooks the agents draw on, rather than a generic prompt.
Launch reAIagents (opens reaiagents.com in a new tab)

Rooms it has stood in.

Government schools in Ranga Reddy district, an industrial training institute, and a trade exhibition floor. Photographs from the visits, not staged shots.

  • Children crowding closely around the robot with a teacher, several reaching toward its screen
  • A full classroom of seated students facing EDU-BOT and a projector screen at the front of the room
  • A trainee taps the robot's chest-mounted touchscreen. The wall display mirrors what the robot is showing: a PLC SCADA training topic, a Listen control, and a box for typing a question
  • Students in school uniform standing either side of EDU-BOT, its screen turned toward them
  • The robot standing at the front of a computer training lab while trainees work at desktop workstations
  • Visitors in conversation beside the robot at a trade exhibition stand

A session, as it ran.

Unedited and silent — one classroom at Kokapet, one demonstration in a training hall.

  • A full class watching EDU-BOT deliver a lesson at Kokapet. Silent clip, 32 seconds.
  • A demonstration in an industrial training institute, trainees and visitors gathered around the robot. Silent clip, 41 seconds.

What the schools said.

Recorded at the schools where EDUBOT has worked. Quoted as given.

  • “With EDUBOT’s arrival, our school saw full attendance. The curiosity and excitement among children were unlike anything we’ve experienced before. Since this humanoid robot can function around the clock, parents feel assured that gaps caused by teacher shortages are being addressed. I am confident that our admissions will double next year, and I strongly encourage every public school to adopt this technology to overcome the teacher shortage.”
    Mr. RamuluHead Master, MPP Public school, Gandipet
  • ”I teach English, but the way EDUBOT explained lessons with animation videos—I have never seen students so engaged. Their enthusiasm for learning was evident. I truly believe EDUBOT can help address today’s educational challenges. This is the need of the hour.”
    Ms. AnithaNGO representative and English teacher
  • “EDUBOT explained the concept of fractions so effectively that I could take a short break. When I returned and wrote problems on the board, students solved them instantly. In that time, I was even able to finish other important work. EDUBOT truly acted like a teacher’s assistant.”
    Mr. SrinuMath teacher, MPP Public school, Gandipet

Press and social

The robot, off our own website.

Coverage and clips published elsewhere. These open on the platform they were posted to — nothing from those sites runs on this page.

Next step

See one in a room.

The machines are built for classrooms, reception halls and hospital corridors — places that are noisy, crowded and unpredictable. A demonstration on site tells you more than any specification sheet.

Request a demoOr watch a recorded session