PhD · MBA · Los Angeles County, California

Pradeep Ravindranath

I have spent thirteen years moving between two rooms — the one where a result has to be defended, and the one where it has to be shipped and paid for. This page is the evidence rather than the claim. Every image below is from work I did, in a domain I had to learn first.

pradeep@dynamworks.com dynamworks.com LinkedIn Google Scholar

Trained to defend a result

PhD, Computational Chemistry: Singapore-MIT Alliance · 2008–2013
Full fellowship. Inverse design of protease inhibitors.
MSc, Informatics (AI): University of Edinburgh · 2006–2008

Trained to run the business

MBA: UC San Diego, Rady School · 2016–2018
Merit scholarship. Technology management.
Since then: Co-founder & CTO, fractional CTO, and founder of an independent practice.

The dates overlap on purpose. I took the MBA in the evenings between 2016 and 2018 while working as a Senior Computer Scientist at USC's Alzheimer's Therapeutic Research Institute — leading the team whose platform helped secure a $70M NIH grant. I did not move from research to business. I have been doing both at once since 2016.

Capability — evidenced 2013 to now

Making a measurement legible

Most scientific results die in the gap between a correct analysis and a figure someone can read. Closing that gap is the work I am most often hired for.

A four-panel scientific figure: a scalp topography heat map with labelled electrodes, a circular clock plot encoding activity periods, a horizontal timeline of temporal windows per electrode, and a scatter plot of activity centre against effect size. — EEG / ERP study · 2026 **Four visual idioms in one figure.** An interpolated scalp topography for *where*; a polar clock encoding for *when*, with 300 ms at twelve o'clock; a per-electrode timeline separating two conditions by fill and hatch; and a scatter placing each electrode by activity centre against effect size. Each panel answers a question the others cannot, and they share one set of electrode labels so a finding can be carried between them. Figure reproduced with the permission of Oksana Shelest.

Close view of the scalp topography: a head outline with nasion marker, an interpolated red-to-blue field, and ten-twenty system electrodes labelled with values in microvolts. — EEG / ERP study · topography detail **Diverging scale, zero pinned at white.** The quantity is signed, so the reader has to see which way it goes — not a rainbow map that hides the sign. Electrodes sit at their 10–20 positions carrying their own values, so the map reads as a field or as a table, whichever the reader needs. Figure reproduced with the permission of Oksana Shelest.

Matrix heat map of recovered ligand-receptor atomic interactions across dozens of structures, coloured red to green, with a companion bar chart. — AutoDockFR · PLoS Computational Biology, 2015 **Recovery of native ligand–receptor interactions**, one row per system, one column per flexible side chain. Docked poses reproduced 79.8% of native interactions on average. A dense matrix has to survive being printed in greyscale and being read one cell at a time; the legend is built for both.

Capability — 2013 to 2016

Seeing structure in three dimensions

Surfaces, meshes and discrete objects competing for the same space, drawn so a reader can see where they agree.

Three stacked panels showing a triangulated wireframe mesh enclosing a ligand, with labelled side chains drawn twice in contrasting colours. — AutoDockFR · 2015 **Apo, holo and docked conformations on one mesh.** The side chains are drawn twice in contrasting colours so the reader sees which residues moved — lysine 33 swings out of the pocket to admit the ligand — without hiding the eleven that did not.

A designed inhibitor in cyan sticks inside a translucent grey molecular envelope, surrounded by pale receptor residues, with attachment points labelled. — Doctoral thesis · 2013 **A designed inhibitor inside the substrate envelope.** Three layers — surface, ligand, receptor — with the surroundings held back in a lighter treatment so the eye reads depth instead of clutter. The inhibitors in this work were designed to stay effective against resistance mutations.

Four panels showing green stick ligands among scattered coloured points, and the same ligands overlaid with large translucent spheres marking predicted feature points. — AutoSite · Bioinformatics, 2016 **Predicted feature points against the measured ligand.** Raw point cloud on the left, clustered prediction as translucent spheres on the right. Colour carries atom type; transparency does the work that would otherwise need a second figure.

Capability — 2020 to 2021

Registering one modality onto another

Autonomous-vehicle sensor calibration, from the Curium years. The domain is far from biology; the problem is not.

A greyscale street scene with parked cars, overlaid with dense red and blue LiDAR points landing precisely on the vehicle bodies. — SPIE Future Sensing Technologies · 2020 **LiDAR points projected onto the camera image after calibration.** The photograph is held in greyscale so the projected points carry all the colour, and the test of the method is whether they land on the cars. Aligning a sparse modality to an image and making the fit visible at a glance is the same problem as putting a measured signal onto an anatomical reference.

Three stacked 3D scatter plots of a car-shaped point cloud: the full cloud, a uniform downsample, and a sparser set of learned feature points. — SPIE Future Sensing Technologies · 2020 **The same object at three densities.** Full point cloud, uniform downsample, learned feature points — on identical axes, so what each reduction throws away is arguable rather than assumed. Every pipeline that has to make large data tractable faces this decision.

Capability — 2016 to now

Research that reached production

A result nobody uses and a product nobody benchmarked are two versions of the same failure. The numbers below are adoption and funding, not accuracy.

$70MNIH grant supported · USC ATRI

60%First-year adoption, medical-coding AI

19 → 2Services consolidated as fractional CTO

24Cross-functional team led

At USC I led the eight-person team building an AI-powered electronic data capture platform for multi-site clinical research, and deployed a medical-coding model that cut coding time from days to hours. As a fractional CTO in 2026 I re-architected a live regulated SaaS platform — nineteen services and four datastores into two domain-aligned services on one system of record — while leading a twenty-four-person team under delivery pressure.

The business side

Built, funded and shipped

Co-founded a venture-backed company, raised against it, and now run an independent practice whose products are in the App Store rather than in a slide deck.

8×Valuation growth in 24 months · Curium

15Engineers built across 3 countries

€1MEureka GlobalStars grant secured

3App Store products live

DynamWorks · live on the site Real-time 3D, state sync, and a grounded assistant in one page. Choosing an avatar updates two further views — the chat panel's header and the launcher in the corner — each holding its own copy rather than sharing one. Behind it the assistant answers only from documents I control: it quotes the published prices, books nothing, and sees no records. Not a demo reel; it runs in the page, on the visitor's machine.

Things you can open right now

Real-time pose detection, 478-point landmark tracking and edge detection, all processed locally — no install, no server, nothing uploaded. Open it on a laptop with a webcam and it runs in the tab.

dynamcv.app.dynamworks.com · dynamworks.com · VisionUX on GitHub

Moving pictures

Watch it run

Two of these are a decade apart. The oldest is research software still being cited; the newest runs in your browser while you read this.

AutoDockFR · recorded at Scripps Research The docking software in use. Recorded when the paper was published. The software is still distributed as part of the AutoDock Suite and still cited — which is the only durability test that counts for research tooling.

DynamCV Playground · screen recording Real-time pose and landmark tracking, in the browser. No install, no server, nothing uploaded — the model runs on the machine watching it. Add media/dynamcv-pose.mp4 and img/dynamcv-poster.jpg to publish this one.

The record

Published, cited, still in use

AutoDockFR — protein–ligand docking with explicit receptor flexibility. PLoS Computational Biology 11(12):e1004586, 2015.

750+ citations

doi.org/10.1371/journal.pcbi.1004586

AutoSite — ligand-binding-site identification and key-atom prediction. Bioinformatics 32(20):3142–3149, 2016.

100+ citations

doi.org/10.1093/bioinformatics/btw367

Self-calibration of sensors using point cloud feature extraction. Proc. SPIE 11525:115250M, 2020. With K. Buyukburc and A. Hasnain.

SPIE

doi.org/10.1117/12.2573123

3D-3D self-calibration of sensors using point cloud data. SAE Int. J. Advances & Curr. Prac. in Mobility 3(3):1369–1377, 2021.

SAE

doi.org/10.4271/2021-01-0086

Continuous Dynamic Calibration for multi-sensor fusion across LiDAR, radar and camera — co-inventor. Published international patent application WO 2022/031226.

patent application

Full record — 9 peer-reviewed publications across computational molecular modelling, clinical machine learning and multi-sensor calibration. Keynote, SAE WCX 2021.

900+ citations

Google Scholar profile

What I do now

I run DynamWorks, an independent practice taking domain-expert product visions into production — fractional CTO engagements, applied-research sprints, and hands-on delivery for healthcare, research and regulated clients. I take the work where the domain matters as much as the engineering.

Pradeep Anand Ravindranath, PhD, MBA
pradeep@dynamworks.com · +1 (858) 366-2815
dynamworks.com · linkedin.com/in/paravindranath

AutoDockFR figures from Ravindranath PA, Forli S, Goodsell DS, Olson AJ, Sanner MF, PLoS Comput Biol 11(12):e1004586 (2015), CC BY 4.0. AutoSite figures (Bioinformatics 32(20), 2016), sensor-calibration figures (Proc. SPIE 11525, 2020) and doctoral thesis figures reproduced by their author. EEG/ERP figure reproduced with the permission of Oksana Shelest.