Karthik Narayan, Developer in Jersey City, NJ, United States
Karthik is available for hire
Hire Karthik

Karthik Narayan

Verified Expert  in Engineering

Machine Learning Engineer and Developer

Jersey City, NJ, United States
Toptal Member Since
May 5, 2020

Karthik is a machine learning expert with over a decade of experience. His career has been punctuated by building large-scale ML algorithms for mission-critical systems including creating ML-based high-frequency trading systems that trade more than 5% of the US equities markets (Hudson River Trading) and creating novel deep-learning-based click-through-rate prediction systems (Google). Karthik also has a Ph.D. in artificial intelligence from UC Berkeley, where he was an NSF and NDSEG Fellow.


Hudson River Trading
PyTorch, TensorFlow, NVIDIA CUDA, Python, C++
University of California, Berkeley
Amazon Web Services (AWS), Ceres, Google, PCL, PyTorch, TensorFlow, NVIDIA CUDA...
Citadel LLC
TensorFlow, Caffe, Python, C++




Preferred Environment

TensorFlow, PyTorch, Unix, Java, Python, C++

The most amazing...

...thing that I've built is a high-frequency trading system with a team of engineers that traded over 5% of the US equities markets.

Work Experience

Algorithmic Developer

2016 - 2018
Hudson River Trading
  • Developed novel high-frequency trading strategies that currently traded more than 5% of US equities; they were primarily liquidity taking and deep learning-based and I was responsible for all stages, from alpha modeling to production implementation.
  • Improved firm-wide profit and loss (P&L) by $XX million/year, in simulation and live trading.
  • Created the HRT AI Labs Fellowship along with the founding partners.
Technologies: PyTorch, TensorFlow, NVIDIA CUDA, Python, C++

Ph.D. Candidate (Machine Learning and Computer Vision)

2012 - 2016
University of California, Berkeley
  • Developed a 3D scanner from scratch using commodity hardware: low-end DSLR cameras, Carmine PrimeSense depth sensors, a turntable, and an arm to hold the devices.
  • Wrote bundle adjustment-based camera calibration software using Google Ceres (used in Google StreetView's calibration) and published the BigBIRD dataset at ICRA 2014. This dataset has been cited 200+ times.
  • Developed novel 3D reconstruction algorithms to produce 3D meshes of objects scanned in the BigBIRD dataset (ICRA 2015). Experiments reveal that scanning accuracy exceeds scanners that cost $100,000+; our scanner costed $3,000+.
  • Developed a novel Levenberg-Marquardt-based 3D mesh coloring algorithm to produce photorealistic 3D scans of the BigBIRD dataset (IROS 2015). User studies indicate that users could rarely distinguish between real images and our model renderings.
  • Developed 3D scanned models that were used by Amazon as part of the Amazon Picking Challenge (2014).
  • Developed the first end-to-end CUDA implementation of t-SNE (a highly popular visualization algorithm in deep learning). Sped up t-SNE runtimes by 60x (ICML 2015:). Generalized t-SNE to allow users to focus on micro and macro data statistics.
Technologies: Amazon Web Services (AWS), Ceres, Google, PCL, PyTorch, TensorFlow, NVIDIA CUDA, Python, C++

Quantitative Researcher Intern

2015 - 2015
Citadel LLC
  • Developed the first deep learning-based alpha models at Citadel Global Quantitative Strategies which currently generate $XX million/year in live P&L. Was responsible for all stages of development, from research to production implementation.
  • Implemented heavily optimized C++ libraries and deployed the alpha model directly into the production trading environment.
  • Implemented an end-to-end research pipeline, allowing a researcher to easily reproduce the alpha models that I had produced during the internship.
Technologies: TensorFlow, Caffe, Python, C++

Research Intern

2012 - 2012
Cornell University
  • Developed novel algorithms to efficiently track photon beam propagation through particle accelerators, particularly the Cornell Electron Storage Ring (CESR) and the Energy Recovery Linac (ERL).
  • Implemented the algorithms in C++ and integrated the system with BMAD, a particle accelerator library developed at Cornell University.
  • Created algorithms that Cornell researchers at CESR and ERL now use to compute mirror placements along the particle accelerator to control photon beam propagation prior to physical implementation.
Technologies: BMAD, Unix, C++

Software Engineer (Google Ads Quality)

2011 - 2012
Google, Inc.
  • Deployed click-through-rate prediction models that generated $XX million/year from Mobile Search Ads traffic.
  • Developed and deployed a novel, large-scale feature selection algorithm responsible for improving click-through-rate prediction performance for Google Mobile Search Ads.
  • Conducted Python code reviews for the full team for Python readability.
Technologies: Python, C++

Software Engineering Intern (Google Help)

2011 - 2011
Google, Inc.
  • Developed a framework for all Google iOS products to handle user-submitted feedback and report iOS app crashes.
  • Created the front-end interface and application using Objective-C and the back end using Java.
  • Internationalized all front-end components in 30+ languages.
  • Began evangelizing the framework towards the end of the internship; now, most Google iOS applications (e.g., search, shopping) employ this framework, allowing millions of users around the world to report feedback for Google iOS products.
Technologies: Java, Objective-C

Undergraduate Researcher

2008 - 2011
Georgia Institute of Technology
  • Developed a multi-agent extension to value iteration, a model-based reinforcement learning algorithm. Published results at AAAI 2011 with Prof. Charles Isbell and Dr. Liam MacDermed.
  • Built a laser-cooled ion trap capable of trapping Ba-138 ions. Produced false-color image of *individual barium ions*. Responsible for mirror and laser calibration and constructing the physical trap. Supervised by Prof. Michael Chapman.
  • Developed a natural language generation framework allowing users to quickly author large amounts of varied text, useful for chatbot applications. Published at AIIDE 2011 with Prof. Charles Isbell and Prof. David Roberts. Press release on Engadget.
Technologies: Python, Java, C++

Software Engineering Intern (Google Search Quality)

2010 - 2010
Google, Inc.
  • Implemented inline satellite/terrain map display in Google Search. For example, try searching for [terrain map of mount everest] on Google.com.
  • Internationalized the satellite/terrain map display functionality across 30+ languages, working with product management and translation teams. For example, try searching for [satelitska karta mount everest-a] on Google.hr; this is Google Croatia.
  • Ran several AB tests to demonstrate that (1) users had positive experiences with this feature and (2) Google ad revenue was not impacted adversely. Verified that the increase in Google search metadata logging was minimal.
Technologies: Unix, C++

ICRA 2014 | BigBIRD: A Large-Scale 3D Database of Object Instances

With a fellow graduate students and two undergraduate students, I built a 3D scanner using commodity hardware.

A fellow graduate student and I collaborated on constructing the physical 3D scanner. Together, we decided on employing a turntable, an arm, and 5 entry-level DSLR cameras and 5 Carmine PrimeSense devices (depth sensors). I developed and wrote a bundle adjustment-based algorithm to calibrate these sensors with respect to each other. Working with another graduate student, we automated the full scanning process—a single click is all it takes to collect all the data associated with a single object: 600 12 MP images, 600 registered RGB-D point clouds, and intrinsic and extrinsic camera poses for each image and point cloud.

I wrote the bundle adjustment algorithm using Google Ceres, a large-scale non-linear optimization library; this library is also used for calibration at Google StreetView. We wrote an end-to-end pipeline in Python that allowed reproducibility of scans and calibration results.

Using this 3D scanner, we published a dataset, BigBIRD, consisting of 125 objects.

We published and presented this work at ICRA 2014.

ICRA 2015 | Range Sensor and Silhouette Fusion for High-quality 3D Scanning

Using the data collected from the BigBIRD dataset, I created a novel 3D reconstruction algorithm capable of scanning objects to within 2 mm of accuracy. The reconstruction algorithm is able to scan objects that are traditionally difficult or impossible to scan by existing scanners; samples objects could be transparent, translucent, or have heavy specularities.

To the best of our knowledge, this is the first paper that fuses visual hull information gathered from RGB cameras with KinectFusion cues from depth sensors; this paper demonstrates that combining these two data modalities can produce 3D scans that outperform scans using only one of these approaches.

At a high level, this approach (1) computes object segmentations from the high-resolution RGB images, (2) computes a visual hull using the segmented, calibrated RGB images, (3) computes a KinectFusion model using the calibrated depth data (Section III-C), (4) refines the original raw depth maps using the visual hull and KinectFusion models and merges these refined depth maps into a point cloud, and (5) forms an object mesh by fusing this merged point cloud with the visual hull.

This work was published at ICRA 2015.

IROS 2015 | Optimized Color Models for High-quality 3D Scanning

Using the BigBIRD dataset and the meshes computed in our ICRA 2014 paper, I constructed a novel algorithm capable of producing photorealistic mesh colorings. Other existing approaches simply compute mesh colorings via averaging: to compute the color of a mesh vertex, simply average the color of that vertex across all images that display that vertex.

Unfortunately, this method doesn't work well because: (1) cameras may not be perfectly calibrated, which may produce heavy artifacts and (2) highly non-Lambertian objects (e.g., soda cans) may produce ghosting effects due to not having the same color observation from all views.

In this paper, I detailed a non-linear least squares problem to solve these issues. Demonstrating that solving this problem naively using Gauss-Newton traditional techniques, I proposed a hybrid coordinate descent and Levenberg-Marquardt algorithm to solve this problem. Ultimately, solving this problem yielded 3D color meshes of very high quality.

I conducted a user study demonstrating that our method outperformed the state of the art substantially; our technique was ultimately able to recover extremely fine details, e.g., lines on barcodes.

Trading More Than 5% of US Equities with Deep Learning-based High Frequency Trading

At Hudson River Trading, I constructed a number of deep learning-based strategies that currently trade more than 5% of the US equities markets. These strategies currently contribute towards a large portion of HRT's P&L. This project was written in Python and heavily optimized C++.

Please contact me directly for questions regarding this project.

Long-term Alpha Modeling with Deep Learning

At an internship at Citadel LLC, I created and deployed the first deep learning-based alpha models at Citadel's Global Quantitative Strategies arm. These strategies produce $xx million/year in live P&L. The models were written in Python and heavily optimized C++.

Please contact me directly for questions.


Python, C++, C, Objective-C, Java, SQL


PyTorch, NumPy, TensorFlow, PCL, Keras


Deep Learning, Deep Reinforcement Learning, Computer Vision, Machine Learning, 3D Scanning, Generative Adversarial Networks (GANs), BMAD, Google, Ceres




Concurrent Programming


Linux, Unix, Docker, Amazon Web Services (AWS), NVIDIA CUDA



2012 - 2016

Master's Degree in Computer Science

University of California, Berkeley - Berkeley, CA, USA

2012 - 2016

Ph.D. in Computer Science (Artificial Intelligence)

University of California, Berkeley - Berkeley, CA, USA

2008 - 2011

Bachelor's Degree in Discrete Mathematics

Georgia Institute of Technology - Atlanta, GA, USA

2008 - 2011

Bachelor's Degree in Computer Science

Georgia Institute of Technology - Atlanta, GA, USA

Collaboration That Works

How to Work with Toptal

Toptal matches you directly with global industry experts from our network in hours—not weeks or months.


Share your needs

Discuss your requirements and refine your scope in a call with a Toptal domain expert.

Choose your talent

Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.

Start your risk-free talent trial

Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.

Top talent is in high demand.

Start hiring