Most recently, at Abacus.AI, he led multiple teams in Generative-AI, forecasting, predictive modeling, vision, and anomaly detection. His GenAI team created 'Dracarys-ver2', which outperforms models like Chat-GPT4O, Claude, and Gemini on coding benchmarks when released.
LLMs link: https://huggingface.co/abacusai/Dracarys2-72B-Instruct
News: https://venturebeat.com/ai/open-source-dracarys-models-ignite-generative-ai-fired-coding/

Sandeep serves as a program committee member for top AI conferences and as a reviewer for leading journals in the field. He has published over 60 research articles in prestigious conferences and journals in computer science. Sandeep has extensive industry experience, having led AI research at Abacus.AI, Amazon, ARM, Teradata, IBM, and Oracle. His research is implemented in production, with applications ranging from controlling robots to designing precision agriculture strategies.

At UCLA, Sandeep served as the president of the Computer Science Graduate Student Association and as the coordinator of the Los Angeles Computing Circle (LACC). As the coordinator of LACC, he extended the STEM programs to several schools in California. He is also an industry expert on AI and an invited speaker for faculty development programs organized by IIT Roorkee.  

Check my books on guiding students for higher education at top universities:
1) https://www.amazon.com/Indian-Students-Guide-U-S-Universities-ebook/dp/B0D2NYFS7W/
2) https://www.amazon.com/Graduate-Admission-Essays-Samples-Statement-ebook/dp/B0D7V66NYP

[LinkedIn]  [Google Scholar] [Youtube] Email (sandha.iitr (at) gmail.com) 

What's Exciting :) 

• Check our Punjab AI Excellence Program (New Batches are available!)

• Checkout my course released by IIT Roorkee: Foundations of AI and Large Language Models (Jan-2026)

• Released new Math & Reasoning benchmarks for LLMs used by Industry (https://livebench.ai/).

• Mango is officially adopted by ARM to design CPUs for Billions of devices (https://github.com/ARM-software/mango).

• My talks at AI programs to schools/universities across India. 

• Selected as Student commencement speaker by UCLA's Henry Samueli School (1 out of a total class of 1000+).

• Program Committee member of AIChallengeIoT-2022.

• Outstanding Mentorship Award by UCLA, 2021. 

• Robust reinforcement learning in collaboration with Amazon AI.

• In-database machine learning is used in production at Teradata [patent].

• REST API for water quality sensing is released along with IBM Research.

• Selected Talks: (i) University of South Carolina [slides], (ii) UCLA [slides],  iii) Statistical Reinforcement Lab, Harvard [slides].

Work Experience

• Punjab AI Excellence (2025 - Present): Sandeep founded and serves as the lead coordinator of the Punjab AI Excellence program.  We are equipping students with the practical AI skills needed to build careers and contribute to India's growing technology landscape. Our program welcomes students from diverse backgrounds, with participants from 200+ cities (80% from Punjab), and 350+ schools—attending live classes to learn AI fundamentals.

• E&ICT Acacemy, IIT Roorkee (2025 - Present): I provide practical Artificial Intelligence training as part of the faculty development programs run by the E&ICT Academy, IIT Roorkee. I have conducted several orientations for top universities in India, exposing more than 2000 professors to practical, state-of-the-art AI skills, including Gen-AI, LLMs, IoT, AI in agriculture, and more.

• Abacus.AI (2022 - Present): I currently work as an Advisor. In the past, I lead several team of engineers/researcher in Gen-AI, forecasting, predictive modeling, vision, and anomaly detection. His GenAI team released "Dracarys-ver2", which not only compete but also outperform models like Chat-GPT4O, Claude, and Gemini on coding benchmarks.
  LLMs link: https://huggingface.co/abacusai/Dracarys2-72B-Instruct
  News: https://venturebeat.com/ai/open-source-dracarys-models-ignite-generative-ai-fired-coding/
  Leaderboard: https://livecodebench.github.io/leaderboard.html

• Amazon (2022): Working on scaling machine learning-based production applications.  

• Amazon (2021): Designed a scalable data-driven tool to handle bottlenecks in production databases. 

• ARM Research (2020): Extended Mango to outperform current state-of-the-art. Mango is used to design commercial CPUs for billions of devices (IoT sensors, supercomputers, smartphones, and laptops).  [code] [slides-1] [pdf-1] 

• ARM Research (2019): Developed Mango, a hyperparameter tuning framework. [code] [demo-video] [slides-2] [pdf-2]

• Teradata Labs (2018):  Enabled distributed deep learning within databases. [patent] [slides-3] [pdf-3]

• IBM Research (2014 - 2016): 

  • Data-driven framework for water quality sensing. [slides-4],  [AAAI tutorial], [pdf-4] [pdf-5]

  • MPI solutions for matrices. [pdf-6] [pdf-7],  [pdf-8], [pdf-9].

• Oracle (2014):  Worked on database recovery features.

• IBM Research (2013):  Distributed matrix solver using MPI. [pdf-10]

 Leadership Roles

• Computer Sci. Grad. Student Association @ UCLA (CS-GSA): Served as the president of CS-GSA from 2017 - 2022.

• UCLA's Summer School for STEM: Coordinator of the STEM school focussed on computer science concepts.  LACC-18 ,  LACC-19 ,  LACC-20 , LACC-21.

• Mobile Dev. Group, IIT Roorkee: Co-founder of the student group on mobile computing research.

Professional Services

• Serving as a reviewer at ICRA, IEEE Big Data, IEEE Sensors, Pattern Recognition Letters, IEEE Internet Computing, Ubicomp, Ubicomm, IEEE Internet of Things Journal, Frontiers of Artifical_Intelligence. 

• Program Committee member of AIChallengeIoT-2022,  ACM SenSys.

• Program Committee member of Serverless-ML-2021, IEEE Big Data Conference

Patent: "Methods and techniques for deep learning at scale over very large distributed datasets," US20200151575A1, 2020.

Selected Publications

• LiveBench: A Challenging, Contamination-Free LLM Benchmark, ICLR 2025 [arxiv link] [pdf]

• Eagle: End-to-end Deep Reinforcement Learning based Autonomous Control of PTZ Camera,  Proceedings of the 8th ACM/IEEE Conference on Internet of Things Design and Implementation, 2023. [pdf] 

• TinyNS: Platform-Aware Neurosymbolic Auto Tiny Machine Learning,  ACM Transactions on Embedded Computing Systems, 2023.

• Inertial Navigation on Extremely Resource-Constrained Platforms: Methods, Opportunities, and Challenges, IEEE/ION Position, Location and Navigation Symposium (PLANS), 2023.

• Neural-kalman gnss/ins navigation for precision agriculture, International Conference on Robotics and Automation (ICRA), 2023. 

• Machine Learning for Microcontroller-Class Hardware: A Review, IEEE SENSORS, 2022. [pdf]

• Auritus: An Open-Source Optimization Toolkit for Training and Deployment of Human Movement Models and Filters using Earables. IMWUT, 2022. [pdf]

• TinyOdom: Hardware-Aware Efficient Neural Inertial Navigation. IMWUT, 2022. [pdf]

• DARTS: Distributed IoT Architecture for Real-Time, Resilient and AI-Compressed Workflows,  Proceedings of the 2022 Workshop on Advanced tools, programming languages, and PLatforms for Implementing and Evaluating algorithms for Distributed systems, 2022.

• Enabling Hyperparameter Tuning of ML Classifiers in Production. CogMI, 2021. [code] [slides] [pdf] 

• Deep Convolutional Bidirectional LSTM for Complex Activity Recognition with Missing Data. Springer, 2021. [code] [slides] [pdf]             

• Sim2Real Transfer for Deep Reinforcement Learning with Stochastic State Transition Delays. CoRL, 2020. [code] [slides] [pdf]

• Time Awareness in Deep Learning-based Multimodal Fusion across Smartphone Platforms. IoTDI, 2020. [code] [slides] [pdf]

• Mango: A Python Library for Parallel Hyperparameter Tuning. ICASSP, 2020.  [code] [slides][pdf]

• Building an open, multi-sensor, dataset of water pollution of ganga basin and application to assess impact of large religious gatherings, IEEE International Conference on Pervasive Computing and Communications Workshops, 2020.

• Exploiting smartphone peripherals for precise time synchronization.  ISPCS, 2019. [code] [slides] [pdf]

• In-database Distributed Machine Learning: Demonstration Using Teradata SQL Engine. VLDB, 2019. [slides] [pdf]

• Radhar: Human Activity Recognition from Point Clouds Generated through a Millimeter-wave Radar. Mmnets, 2019.[code][slides] [pdf]

• Enabling Edge Devices that Learn from Each Other: Cross Modal Training for Activity Recognition. EdgeSys, 2018. [code] [slides] [pdf]

• Deep convolutional bidirectional LSTM based transportation mode recognition, Proceedings of the 2018 ACM international joint conference and 2018 international symposium on pervasive and ubiquitous computing and wearable computers, 2018.

• Data hub architecture for smart cities. In Proceedings of the 15th ACM Conference on Embedded Network Sensor Systems, 2017.

• Comparison of discrete and equivalent continuum approaches to simulate the mechanical behavior of jointed rock masses, 1st International Conference on Energy Geotechnics, 2016.

• Modeling of discrete intersecting discontinuities in rock mass using XFEM/Level set approach,  Energy Geotechnics: Proceedings of the 1st International Conference on Energy Geotechnics, 2016.

• A multi-sensor process for in-situ monitoring of water pollution in rivers or lakes for high-resolution quantitative and qualitative water quality data, IEEE Intl Conference on Embedded and Ubiquitous Computing, 2016.

• Blue water: A common platform to put water quality data in India to productive use by integrating historical and real-time sensing data, IBM Research Report, 2015

• Multiple intersecting cohesive discontinuities in 3D reservoir geomechanics, ARMA US Rock Mechanics/Geomechanics Symposium, 2016.

• Discrete modeling of multiple discontinuities in rock mass using XFEM, ARMA US Rock Mechanics/Geomechanics Symposium, 2015.

• Crowd-pan-360: Crowdsourcing based context-aware panoramic map generation for smartphone users, IEEE Transactions on Parallel and Distributed Systems, 2014.

• Mobile health application for early disease outbreak-period detection,  IEEE 16th International Conference on e-Health Networking, Applications, and Services (Healthcom), 2014.

• XFEM Formulation of Geomechanics Problems with Multiple Intersecting Discontinuities, 10th Biennial International Conference & Exposition on Petroleum Geophysics, 2013.

 Fundamental ML Skills

• ML in production, Distributed Analytics, Large-scale Databases/Infrastructure [Paper-1] [Paper-2] [Paper-3] [Patent]

• Multimodal Sensing/Fusion, Activity Recognition, Tracking, Mobile Computing, TinyML, IoT/CPS [AAAI tutorial] [Paper-4] [Paper-5] [Paper-6] [Paper-7] [Paper-8] [Paper-9] [Paper-10] [Paper-11]

• Reinforcement Learning, Computer Vision, Modelling, Simulation [Paper-12] [Paper-13]

Active Research Projects

Research Themes: Develop artificial intelligence systems by advancing research in sensing, algorithms, scalable learning, and lightweight computation for edge devices. My focus is to enable critical real-world applications in the areas of water monitoring, semiconductor designing, precision agriculture, and ocean monitoring.

1. Multimodal Sensing and Machine Learning for Complex Data Analytics
   In order to advance critical applications in water quality monitoring and human activity prediction, we are working on conducting fundamental research in real-time sensing, data analytics, and knowledge extraction. For water quality monitoring in collaboration with IBM Research and other research groups (https://researcher.watson.ibm.com/researcher/view_group.php?id=6924), we are developing a groundbreaking multi-sensor process for in-situ data collection without the need for laboratory analysis. We are also working on researching novel machine learning for human activity monitoring that can enable a new era in healthcare, fitness, and customized AI.
   
   

2. Developing Scalable Machine Learning Techniques
   In this project, we are focussing on developing scalable machine learning to address critical applications like designing semiconductor chips and handling distributed databases. The complexity of modern semiconductor chips, with billions of transistors, has made manual verification of new chip designs nearly impossible. To automate the chip verification, we are developing new machine-learning research which is officially adopted by Arm (https://github.com/ARM-software/mango). Another direction that we are exploring in collaboration with Teradata is to allow machine learning models to be trained directly within the distributed database itself.
   
   
   

3. Enabling Lightweight Artificial Intelligence for Edge Devices
   We are surrounded by edge devices including small robots, smartphones, smartwatches, airpods, cameras, speakers and wristbands, etc. Despite the potential of these edge devices to revolutionize various applications such as agriculture, ocean monitoring, and rescue operations, the limited memory and computation power of these devices has hindered the deployment of complex machine-learning algorithms. To address this, in this project, we are working on developing tiny machine-learning models. Our vision is to enable mobile robots equipped with various sensors (e.g., vision, audio, soil moisture, inertial measurement unit), and actuators (camera control, fertilizer dispenser, vehicle controls) to aid agricultural practices, ocean monitoring, and rescue operations in real-time. 

More Details

1-Scalable and Robust Machine Learning

2-Deep Reinforcement Learning based Active Tracking

We introduce a novel simulator developed over Unreal Engine controlling objects in photo-realistic virtual worlds. The simulator enables the training of deep-RL policies for tracking, outperforming traditional object detectors and Kalman baselines. 

Sim2Real: We show the transfer of models trained in simulation to the real world.  Paper: Accepted in IoTDI-23.

3-Robust Delay-aware Deep Reinforcement Learning

4-Deep Convolutional Bidirectional LSTM for Complex Activity Recognition with Missing Data, Springer (2021).

• Swapnil Sayan Saha*, Sandeep Singh Sandha*, Mani Srivastava.

How does ML behave with runtime uncertainties? Code  PDF  Video

(3rd in the Cooking Activity Recognition Challenge out of 78 teams.)

5-Time Awareness in Deep Learning-Based Multimodal Fusion Across Smartphone Platforms, IoTDI -2020

Tell me how accurate is my smartphone's time? Does it impact my ML applications?

6-Exploiting Smartphone Peripherals for Precise Time Synchronization, ISPCS-2019

What is the best possible timing accuracy on modern smartphones? oh! yes, it is in microseconds.

7-Radhar: Human Activity Recognition from Point Clouds Generated through a Millimeter-wave Radar. Mmnets, 2019

We present end-to-end deep learning for sparse point clouds from Radar and introduce a new open-source dataset. 

code slides pdf

8-In-database Distributed Machine Learning, VLDB-2019 

• (In collaboration with Teradata Labs)

• Can we train deep-RL models within a database itself without moving data?

9-Heliot: Hybrid emulation of learning enabled IoT systems. 

Sandeep Singh Sandha, Mani Srivastava.  Code

Best Demo award at IoTDI 2019

Processing health data in real-time using Apache Kafka and Spark (UCLA)

Project Page  and Git Code  Project Report 

Distributing TensorFlow graph computation (UCLA)

Project Page Project Presentation  GitHub 

Real-time query interface for large datasets using Apache Spark (UCLA)

 Slides , Project Report , GitHub 

Numerical solution of PDE using distributed computing (IBM Research)

Smartphone based remote health monitoring system (IIT Roorkee)

Context-aware panoramic map generation using opportunistic crowdsourcing (IIT Roorkee)

CP360 generates a fully-tagged 360 degree panoramic map of the surroundings of a querying user using crowd-sourced images, audio trails, object tags, and raw location data collected by smartphones in an opportunistic manner.  IEEE TPDS 2014. 

IBM national technical challenge, 2013 (IIT Roorkee)

This was project done as part of IBM National Technical Challenge (NTC 2013) to predict crime prone areas and suspects. Our work was accepted among the top five teams in finals. 
Slides (Slides are released as they were!)