Sai Ramana Kiran

I am a Distributed Training/Inference Engineer II at AWS.

At Quidient, as a Senior Algorithms Engineer, I developed 3D reconstruction algorithms at the intersection of computer graphics, computer vision and deep learning. During my master's at WPI, I worked on 3D computer vision on edge devices for tiny aerial robots in the PeAR Lab, advised by Prof Nitin Sanket. From 2018 to 2022, I was a senior C++ developer at Quantitative Brokers, building latency-sensitive trading systems and fast execution algorithms.

I hold a BS/MS in Aerospace Engineering from IIT Madras, where I worked on collaborative navigation for aerial and ground robots in Prof Ranjith Mohan's lab.

CV / Email / GitHub / Google Scholar / LinkedIn


Distributed Training/Inference Engineer II AWS Jan 2026 - Present	Senior Algorithms Engineer - CV/CG/ML Quidient May 2024 - Jan 2026	Computer Vision and Deep Learning Intern Torc Robotics Aug 2023 - Dec 2023	Perception and Localization Intern Nokia Bell Labs Summer 2023 Spring 2024	M.Sc Robotics WPI 2022-2024	Senior C++ Developer Quantitative Brokers Jul 2018 - July 2022	B.Tech/M.Tech Aerospace Engineering IITM 2013-2018

Research

I'm interested in math, computer vision, machine learning, optimization, graphics and their applications to robotics.

EdgeFlowNet: 100FPS@1W Optical Flow For Tiny Mobile Robots

Sai Ramana Kiran, Rishabh Singh, Manoj Velmurgan, Nitin Sanket

arXiv / GitHub / project page / video

We present an approach to achieve 100FPS optical flow inference speeds and demonstrate on real world robot autonomy experiments. Our work has been accepted to IEEE Robotics and Automation Letters (RA-L) and will be presented at ICRA 2025.

Aerial-Ground Robotic system for Terrain estimation and Navigation

Sai Ramana Kiran, Ashutosh Kumar, Ranjith Mohan
ICC, 2019

IEEE Paper / Aerial vehicle stack / Ground vehicle Stack

We show autonomous aerial vehicle and ground vehicle collaborate while performing task of unknown rough terrain estimation and navigation.

Vibecoded apps

Products I designed and built end to end in conversation with an AI coding agent, and use every day.

Timeslice

A keyboard-first time tracker for the Mac and iPhone

product page & live demo / code

Switch tasks like ⌘-Tab, see where the day went, and never babysit a stopwatch. Fuzzy focus, goals that know what week it is, one timer across every device, and sync through your own Google Drive — no account, no server. v1.0, built with Claude Code.

Projects

These include coursework, side projects and unpublished research work.

	Enhancing Multi view structured Bird’s-Eye-View Traffic Scene Understanding from Onboard Images slides / project page Enhanced the original STSU’s transformer architecture to use multi-view onboard images using enhanced positional encoding to construct better lane lines
	Stereo Visual Inertial Odometry project page Implemented Multi-State Constraint Kalman Filter based Visual Inertial Odometry
	Point Cloud Semantic Mapping project page A pipeline to combine the data from LiDAR and Camera to obtain semantic painted point cloud.
	Neural Radience Fields (NeRF) project page Implemented a full pipeline of NeRF
	Structure from Motion (SfM) project page Simultaneously reconstructed 3D scene (Mapping) and extracted camera pose (Localization) from given camera correspondences using (Non)Linear triangulation, (Non)Linear PnP and Bundle Adjustment (BA) pipeline
	AutoCalib: Zhang's Calibration module project page Implemented Zhang’s camera calibration by nonlinear optimization of intrinsics and extrinsics
	Face Swap project page Faceswap pipeline using Delaunay triangulation and Thin plate splines based warping and poisson blending in a video
	AutoPano: Stitching assorted images project page Created a panorama by stitching images using homography estimated from traditional (ANMS, RANSAC feature points) and Deep learning (Supervised and Unsupervised HomographyNet) methods