Shivendu
Kumar
I build machines that have to work out where they are and what they are looking at — a racing car with no driver, a drone mapping a mine where GPS cannot reach, a camera sorting waste on a moving belt. Most of my useful work has been figuring out why something failed when the obvious explanation was wrong.
Where I have worked
three placementsNew Delhi
Shoonya Recycling
- An automated sorting line was missing throughput and nobody could say why. I owned the diagnosis and the fix, working between engineering and operations.
- My first assumption was the model needed more training data. It did not. Comparing misclassifications against what physically happened on the belt showed the failures clustered on motion blur and underexposure — an input problem, not a model one.
- Fixed with CLAHE contrast enhancement and a move from a CNN to a vision transformer, whose global attention holds up where local texture is unreliable.
Mumbai
Vazirani Automotive
- Designed an axial fan compressor doing two jobs at once: generating downforce and feeding forced-air cooling to the battery pack.
- Ran the baseline CFD in ANSYS — a blade passage meshed in TurboGrid to a 3 µm first cell, then the 11-blade, 12,000 rpm stage solved in CFX with SST k–ω at 0.07 kg/s.
Jodhpur
IIT Jodhpur — School of AI & Data Science
- Fused camera, ground-penetrating radar and IMU streams from UAV trials into one perception pipeline, handling three different sampling rates and failure modes.
- Built the evaluation harnesses that made detection performance legible across terrain, turning an inconclusive result into a roadmap the team acted on.
Things I built
sixF1/10th autonomous racing stack
Nobody assigned this one. A full autonomous racing stack on Linux: 2D LiDAR SLAM with particle-filter localisation, closed-loop Pure Pursuit tracking, and a reactive gap-finding controller that picks a line through obstacles at speed.
Read the codeSearch-and-rescue UAV
Two sensing modes for finding people: YOLOv5 on the surface, simulated ground-penetrating radar below it. 84.7% detection confidence across 50+ runs. The part worth talking about is not the number — it is knowing which conditions made each sensor lie.
Read the codeMapping a mine with no GPS and no light
Underground tunnels defeat visual localisation: no ambient light, and brick walls that repeat every few metres so every frame looks like the last. An unscented Kalman filter fusing 100 Hz IMU with sparse visual features and LiDAR cut positional drift 82.4%, down to 0.07 m over a 60 m gallery. I am first author on the paper.
Industrial autonomous mobile robot
A 500 mm cube carrying 20 kg at 1 m/s, drivetrain sized from first principles to a 2.0 safety factor. 6 mm aluminium base plate, an ABS mezzanine isolating the compute from motor vibration, and a TS35 DIN rail so standard components mount without custom brackets.
Why half a good idea made things worse
Owl-inspired serrations on a UAV propeller, three geometries, transient large-eddy simulation. Full-span serration cut noise 3.87 dB. Serrating only half the blade made it 1.87 dB worse — and the cause was not the serration but the junction between treated and untreated sections. The transition was wrong, not either half.
Read the studyHealth Atlas — EY Techathon 6.0
A healthcare analytics product taken from research to working software in a competition sprint. Market and competitor research first, then requirements and wireframes, then a multi-agent pipeline with an NLP layer that turns unstructured reports into ranked KPI dashboards someone non-technical can actually read.
Read the codePapers
twoAI-Enabled Autonomous Mapping in Underground Environments
Shivendu Kumar (first author), Dr. Golak Bihari Mahanta. A hybrid unscented Kalman filter and safe-flight-corridor architecture for tracking miners where satellite positioning does not reach. Funded by the TEXMiN Chanakya Undergraduate Fellowship.
Modeling and Simulation of UAV-Based Search and Rescue
Shivendu Kumar, Dr. Bhivraj Suthar. A unified perception, planning and control architecture in Webots for autonomous victim detection. I contributed the data analysis and the quantitative evaluation of detection performance.
What I work with
tools- Languages
- Python (Pandas, NumPy, SciPy, Scikit-learn), C++, SQL — MySQL and PostgreSQL. Git, Linux as daily environment.
- Robotics
- ROS and ROS2, Gazebo, Webots. 2D LiDAR SLAM, AMCL localisation, unscented Kalman filtering, sensor fusion, PID and closed-loop control, A* planning.
- Vision and ML
- PyTorch, OpenCV, convolutional networks, YOLOv5, vision transformers, CLAHE, multi-agent pipelines, NLP extraction from unstructured text.
- Simulation
- ANSYS Fluent, ANSYS CFX, TurboGrid. LES-WALE turbulence modelling, Ffowcs Williams–Hawkings acoustic analogy, aeroacoustics, rotor dynamics.
- Design
- SolidWorks, Autodesk Fusion 360. Part and assembly modelling, drivetrain sizing, material and mounting selection, structural analysis.
- Analysis
- Root-cause investigation, experiment design, statistical analysis, KPI dashboards in Power BI and Streamlit, advanced Excel.
- Languages spoken
- English, Hindi.
Honours and roles
selectedChanakya Undergraduate Fellow
TEXMiN Foundation, IIT (ISM) Dhanbad. One of a small number of undergraduates nationally funded to define and run an independent research project.
First place and Best Speech, UNESCO-WWDR Model UN
For a data-backed technical strategy on antimicrobial resistance, argued before an international panel.
Team Sub-Lead, ISRO URSC Challenge
Led technical contributions and coordinated hardware integration and sensor validation through to the national elimination round.
Semifinalist, EY Techathon 6.0
Among the top teams nationwide from more than 115,000 registrations.
Finalist, EU-India Ideathon
Marine plastic pollution challenge, among 150+ international participants.
Office Bearer, Hackslash Developers Club
NIT Patna. Previously on the content team. Also Event and PR Lead at ISIE NITP SRA, sponsorship lead for ByteVerse, and organiser of the Smart India Hackathon presentation round.
Ranked 25 nationally, Naukri Campus EROH
From more than 110,000 applicants. Also top five at Material Spark 2025, IIT Patna.