Research that leaves the lab
Dreamroot is an AI company built on fifteen years of research and engineering across India and Germany — doctoral work at Leibniz University Hannover, papers at the field's top venues, and mission-critical software delivered for defence and industry.
Dr. Ashutosh Sao
The work started at Bharat Electronics in Ghaziabad, leading various software modules for Indian Navy and Air Force. Five years of building real-time software where failure was not an option.
Then Germany: a master's at the University of Bonn on detecting objects and their states, a year at Gestigon adapting object detection to depth cameras for the automotive industry, and a doctorate at Leibniz University Hannover on data-driven methods for spatio-temporal prediction, supervised by Prof. Dr. Wolfgang Nejdl.
Alongside it, five years of applied research at the L3S Research Center — charging-demand forecasting on a joint project with Volkswagen, and a graph learning framework for detecting illicit activity in trade data from 32 countries.
Dreamroot exists to turn cutting-edge AI research into solutions for real-world problems. The methods are published and the numbers are measured, but the systems that need them run on real equipment, on private networks, on hardware bought years ago. We build for that.
- PhD Computer ScienceLeibniz University Hannover, Germany2020 · 2025
- MSc Computer ScienceUniversity of Bonn, Germany2015 · 2018
- BTech Computer Science and EngineeringNIT Raipur, India2006 · 2010
Data-Driven Methods for Spatio-Temporal Prediction
Where the work has been presented and delivered
Research is judged in public. Ours has been presented, reviewed and organised at venues across three continents.
- 01 · India
CRL-BEL, Ghaziabad
Five years at the Central Research Laboratory of Bharat Electronics, on systems for the Indian Navy and Air Force.
- 02 · USA
IEEE ITSC 2021
Charging station occupancy forecasting, presented online while travel was closed by the pandemic.
- 03 · Japan
PAKDD 2023
Presented MetaCitta, a meta-learning framework for spatio-temporal prediction across cities and tasks.
- 04 · Greece
ESSAI & ACAI, 2024
A week of advanced coursework and exchange with researchers on the European Summer School on Artificial Intelligence 2024.
- 05 · Spain
ECAI 2024
Presented ST-SampleNet, a spatially constrained transformer that cut computational cost by 40%.
- 06 · Germany
WSDM 2025
Part of the organising team in Hannover, the city where the doctoral and applied research was done.
- 07 · Italy
CIKM 2026
METrIC-G, on imbalanced classification in transaction graphs, accepted for Rome in November 2026.
- 08 · India
Durg, Chhattisgarh
Where Dreamroot is based, in the middle of India's steel belt and close to the plants we build for.
The methods are on the record
- CIKM 2026 · Core A
METrIC-G: Multigraph Enrichment and Time-Aware Message Passing for Imbalanced Classification in Transaction Graphs
Trade data, 32 countries - ECAI 2024 · Core A
Spatially Constrained Transformer with Efficient Global Relation Modelling for Spatio-Temporal Prediction
−40% compute, −1% accuracy - PAKDD 2023 · Core A
MetaCitta: Deep Meta-Learning for Spatio-Temporal Prediction Across Cities and Tasks
94.7% less training time - KDD 2022 · Core A*
Reinforcement Learning-based Placement of Charging Stations in Urban Road Networks
Up to 97% shorter waits - IEEE ITSC 2021 · B1
Deep Information Fusion for Electric Vehicle Charging Station Occupancy Forecasting
593 charging stations
Fifteen years of systems that had to keep running.
- 2025 ·Dreamroot · Durg, India
Custom AI systems for industry and enterprise, engineered to run on your premises.
- 2020 · 2025L3S Research Center · Germany
Charging demand forecasting on the D-E-MAND project with Volkswagen, then graph learning for illicit trade detection on the ATTENTION! AI project. Supervised master's students throughout.
- 2019Gestigon GmbH · Germany
Object detection on depth images for in-cabin automotive sensing, deployed through the TensorFlow C API.
- 2017 · 2018University of Bonn · Germany
Detecting objects and their states simultaneously, on architectures based on Faster R-CNN.
- 2010 · 2015Central Research Laboratory, Bharat Electronics · India
Led the diagnostics module for the Indian Navy's Combat Management System and the network management system for the Indian Air Force.