Hironmoy Roy Rudra

ML systems from data to deployment: PyTorch and PEFT pipelines, annotation stacks with SAM and YOLO, TensorRT and TensorFlow Lite at the edge.

BSc Computer Science, BRAC University, Dhaka. Expected graduation January 2027. Available remotely from January 2027.

AI/ML Software Engineer CV (PDF) hironmoy5458@gmail.com github.com/Rudra-roy linkedin.com/in/hironmoy-roy-rudra

About

Computer Science undergraduate at BRAC University who builds ML systems from data to deployment. Built PyTorch training and PEFT pipelines (LoRA, AdaLoRA, TSDLoRA) with Optuna hyperparameter search; dataset collection, labeling and semi-automatic annotation stacks using SAM and YOLO-family detectors; and edge inference with TensorRT on a competition robot and TensorFlow Lite on Raspberry Pi with post-training quantization.

Comfortable in Python, C++ and TypeScript, from CLI and web-portal tooling to ROS 2 nodes, with tested, reproducible pipelines.

Current work

LoRA rank allocation

Under review

Transactions on Machine Learning Research · first author

A matched-control study of probe-based rank allocation for LoRA across five seeds, three decoders and eight tasks, reporting a negative result honestly.

Adapter rank allocated across transformer layers Twelve transformer layers drawn as columns. A pale uniform allocation sits at the same height in every column with a dashed budget line across it; the searched allocation rises in each column to a different height, some above the line and some below, while the total stays on budget. budget layer rank
Rank-allocation sketch drawn for this page: adapter rank per transformer layer against a shared budget, with a uniform allocation behind it; not a figure from any manuscript.

DhakaSense annotation

Under review

semi-automatic 3-D labelling behind an ICRA 2027 submission · second author

Semi-automatic annotation pipeline that combines a COCO detector, a regionally fine-tuned detector and foundation-model segmentation lifted into the LiDAR point cloud, over 14 h 35 min of dense Dhaka traffic.

COVE dataset

In preparation

NeurIPS 2027 Datasets & Benchmarks track · collaboration

Event plus RGB captures of rule-breaking hazards across five seeded industrial environment families with exact label geometry and QC gates.

Projects

PEFT multi-seed retraining suite (TSDLoRA, LoRA, AdaLoRA)

2026

  • PyTorch
  • peft
  • Optuna
  • GLUE
  • Wilcoxon tests

Three-stage pipeline behind the thesis experiments: Optuna/TPE hyperparameter search, multi-seed retraining of 3 methods × 3 backbones × 8 GLUE tasks (72 combinations, 5 seeds each) through a CLI or a web portal, and an analysis stage producing mean ± std tables, single-signal ablations and Wilcoxon tests.

DhakaSense capture and annotation stack

Jul – Sep 2026

  • ZED SVO2
  • RealSense
  • SBG Ellipse-D INS
  • RTKLIB PPK
  • CVAT
  • NFS

In-vehicle recording stack with supervisor and watchdog, PPK-corrected INS trajectories, a remote export pipeline that hashes hundreds of GB where they live, and a semi-automatic 3-D labelling pipeline lifting foundation-model segmentation into the LiDAR cloud. Private repository.

hifi-environment-builder (hifienv)

Public code

Oct 2026

  • Python
  • Unreal Engine 5
  • Cosys-AirSim
  • Isaac Sim
  • USD
  • Blender

Python toolkit and Claude Code skill that builds label-correct, sensor-clean, licence-clean, reproducible simulation environments for Unreal Engine 5 / Cosys-AirSim and Isaac Sim / USD from CC0 and CC-BY assets with provenance; parity, null-test and licence gates; 289 tests.

Four renders of the same warehouse aisle with placed hazards: dropped pallets, a drone, a forklift and a person between the racking
One warehouse aisle, four hazard placements.
Five lighting variants of one warehouse aisle, each labelled with its variant name
Five lighting variants of the same aisle.

EchoMind and my_bot

Public code

Sep 2024 – Sep 2025

  • TypeScript
  • ROS 2
  • Nav2
  • URDF

EchoMind: a TypeScript wrapper for locally hosted vision-language models with an API-driven, asynchronous client-server loop. my_bot: a ROS 2 mobile robot with full URDF, Nav2 and LiDAR/IMU SLAM in simulation.

evs-bench

Jul – Sep 2026

  • Python
  • ESIM event synthesis
  • YOLO11
  • SORT
  • AirSim

Event-based surveillance benchmark tooling: corrected ESIM event synthesis, class-agnostic RGB and event detector baselines, dt-aware SORT tracking and a time-to-first-detection metric; now the capture infrastructure behind COVE. Private repository.

Publications and manuscripts

Listed by title only: no PDFs, and no figures, from manuscripts under review or in preparation. No author lists for unpublished work.

Matched Controls for Probe-Based LoRA Rank Allocation: A Negative Result for One Allocator under a Rank-Coupled Initialization

Under review

Transactions on Machine Learning Research, first author

Five-seed, budget-matched comparison of a probe-based rank allocator against uniform-rank LoRA, LoRA and GoRA on eight commonsense tasks across three decoders plus 11 of 24 GLUE cells; the allocator trails the matched control in 26 of 35 cells.

DhakaSense: A Multimodal Driving Dataset and Benchmark for Dense, Heterogeneous Traffic in Dhaka

Under review

IEEE ICRA 2027, second author

14 h 35 min across 175 scenes from a Livox Mid-360, stereo depth, six RGB cameras and post-processed GNSS/INS, with CNG auto-rickshaws and cycle rickshaws as 3D classes and every metric reported per traffic-density level.

CERTUS: Certified Terrain Traversal Under Sensing Uncertainty, with Certification-Driven Active Perception

In preparation

IEEE/RSJ IROS 2027, first author

A rover drives only on ground it has certified: step, slope and body clearance pass with a distribution-free margin on the map's own uncertainty; a decidability rule predicts which depth sensor can certify which rover; when nothing ahead is certified, active perception picks the viewpoint that resolves the most blocking cells per second. Validated in Isaac Sim before deployment on a real rover.

COVE-Sim: counterfactual, recipient-specific communication for event-camera UAV teams (working title)

In preparation

IROS 2027, first author

The first collaborative-perception system whose shared representation is event-derived, with a counterfactually supervised, recipient-specific gate deciding what crosses the link; simulation study in Unreal Engine 5.8 / Cosys-AirSim.

COVE multi-UAV hazard detection and avoidance dataset (working title)

In preparation

NeurIPS 2027 Datasets & Benchmarks track, collaboration

Five industrial environment families with seeded layouts, exact label geometry, QC gates and rule-breaking hazard speeds, captured with event and RGB cameras.

Experience and honours

  • Education

  • BSc Computer Science, BRAC University, DhakaSep 2022 – Jan 2027 (expected)
  • Higher Secondary Certificate, Science, National College of Education, Narsingdi2019 – 2021
  • Undergraduate thesis: TSDLoRA: Prioritizing Task-Specific Layers for Efficient Low-Rank Adaptation in Encoder-Only Transformers2026
  • BRACU Mongol-Tori (University Mars Rover Team)

  • Co-AdvisorNov 2025 – present
  • Autonomous Sub-Team LeadSep 2024 – Aug 2025
  • Control Systems EngineerFeb – Aug 2024
  • ContributorNov 2023 – Feb 2024
  • Elsewhere

  • Kibo Robot Programming Challenge, 6th edition (JAXA): 6th place, national roundAug 2025 – Feb 2026
  • First Global Challenge Team Bangladesh: Technical MentorSep 2024 – Feb 2025
  • ROBO hobby: AI Engineer, edge computing for agricultureOct 2024 – Jan 2025
  • Robotics Club of BRAC University: Student Instructor, ROS, Gazebo and WebotsOct – Nov 2025
  • Honours

  • 8th place, University Rover Challenge 2025 (The Mars Society, Utah)
  • 21st place, University Rover Challenge 2024
  • VC's List ×4 and Dean's List ×3, BRAC University

CV

The CV and this page say the same things; the CV has the detail.