Neelay RanjanGenerative modeling for domains where labels are scarce and mistakes are expensive.

NASA Ames Research Center · Regenstrief Institute · Purdue University

I build diffusion and energy-based models. I’m first author on a Regenstrief paper, in preparation, on segmenting blood vessels from very few labels. At NASA Ames I built a diffusion model that reroutes flights around closed airspace, and now I work on speech recognition for air traffic control. I’m on a gap semester from Purdue, applying to master’s programs for fall 2027.

press a face to sample it

Press a face and a diffusion model with 1.3M parameters, overfit on three photos of me, samples it in your browser from noise in 25 steps. Another face starts from the current picture.

IN PREPARATIONSupplementary materialdiffusion trajectories · MAE vs I-JEPA · a day of synthesized flight plans · SDEdit on MNISTSeptember 2026

**Table 1.** What the page shows, and where.
Finding	Result	Shown in
Label efficiency	0.882 Dice from 16 labeled angiograms, ahead of five baselines in 25 of 25 paired runs (paper in preparation)	Figure 1
Failures	0.32% of predictions under 0.5 Dice, against 5.4% for SAM	Figure 2
Surgeon time	~75% faster corrections, measured with one surgeon	Research
Airspace rerouting	A diffusion model reroutes flights around closed airspace with no retraining	Figure 3
Chess engine	Roughly 1900-2200 Elo with search on a Raspberry Pi, from a 553 KB model	Figure 4
Titan lander	An autopilot that plans inside a diffusion model of its own flight	Figure 5

Research

At Regenstrief I trained a diffusion model to predict the vessel mask directly. From 16 labels it reaches 0.882 Dice, ahead of five baselines in all 25 paired runs. Trained to predict noise, the same architecture stays near 0.23.

The same model made a surgeon ~75% faster at correcting its output.

paperRanjan, Dev, Gonzalez. “Bootstrapping surgeon labeling campaigns with x0-diffusion: label-efficient vessel segmentation of catheter-based angiograms.” In preparation.

studyMeasured with one vascular surgeon, my coauthor Dr. Andrew Gonzalez.

log x axis. budgets 16 (5%), 32 (10%) and 80 (25%) labels of the 320 training images.

Figure 1. Mean test Dice over every seed and fold. At 16 labels x0-diffusion leads the next best, zero-shot SAM, by 0.047. By 32, DeepLabV3 and ResNet-UNet catch up. ε-diffusion (about 0.23) is off the chart.

dataThe angiograms in Figures 1 and 2 are from the public pelvic-iliac benchmark.

scopeLabel efficiency is the claim. With more labels, some baselines catch up.

Pelvic-iliac angiogram from the segmentation benchmark, test image 383 — **Figure 2.** At 16 labels, 0.32% of x0-diffusion’s predictions fall below 0.5 Dice, against 5.4% for SAM and 13.5% for ResNet-UNet. x0-diffusion’s mean is only a few points ahead, but it almost never fails outright. Each slider stop shows the image whose SAM Dice is nearest the cursor. SAM is stochastic, so its mask can score differently from its recorded run.

**Figure 2.** At 16 labels, 0.32% of x0-diffusion’s predictions fall below 0.5 Dice, against 5.4% for SAM and 13.5% for ResNet-UNet. x0-diffusion’s mean is only a few points ahead, but it almost never fails outright. Each slider stop shows the image whose SAM Dice is nearest the cursor. SAM is stochastic, so its mask can score differently from its recorded run.

I also led the PCB design team for a helical stimulation antenna for Alzheimer’s therapy, presented as an oral at IEEE MWSCAS 2026.

referenceF. Perez, J. Morisaki, H. Kanakri, M. Rizkalla, et al. (incl. N. Ranjan), “Helical Antenna for Electromagnetic Field Stimulation in Alzheimer’s Disease Therapy,” IEEE MWSCAS 2026 (oral).

NASA Ames Research Center

On SLAAC (space launch and air/airspace coordination) I trained a diffusion model on FAA radar tracks to reroute flights around closed airspace. It applies the airspace while it samples, so a new closure needs no retraining.

Since August I’ve been on SHIFT. My speech recognizer for air traffic control makes fewer errors than the Whisper system in use, and a generate-and-rank parser turns its transcripts into a maneuver database. I’m writing the parser up with NASA for AIAA AVIATION.

slaacOn the SLAAC poster, nearly every final reroute kept every leg the full buffer from the airspace, with small detours close to the geometric optimum.

dataPublic data where I could, at some loss of quality. NASA’s airspace file isn’t public, so the airspace is the FAA’s Special Use Airspace layer. With no historical route database, the filed routes are my model’s.

differencesMy SLAAC models are integrated into the simulation software NASA and the FAA use to evaluate future air traffic management. This demo samples the same weights in fewer steps on public airspace, in seconds on my laptop and longer with more airspace or every flight at once. Fewer reroutes keep the full buffer than on the poster, though almost none cross the airspace.

Live systems

Both demos below run the trained models in your browser.

Experience

since Aug 2026NASA Ames · SHIFTSpeech recognition for air traffic control, a generate-and-rank parser from transcripts to maneuvers, and a synthetic speech corpus about 20x the recorded data.active

May-Aug 2026NASA Ames · SLAACDiffusion rerouting around closed airspace, and a flight plan language model that writes a day of about 44,000 flights.complete

2024 to presentRegenstrief Institutex0-diffusion vessel segmentation and synthetic angiograms, with Shantanu Dev and Dr. Andrew Gonzalez.active

expected May 2027Purdue UniversityB.S. in Artificial Intelligence, Intelligent Control & Systems concentration, math minor. John Martinson Honors College, Dean’s List, 3.68 GPA.gap semester

2026Alzheimer’s stimulation antennaLed the PCB design team on a helical antenna for electromagnetic field stimulation, presented as an oral at IEEE MWSCAS 2026.complete

2025Davinci WearablesAn agentic vision pipeline that estimates nutrition from meal photos, under 15% error.complete

2024-2025V2X aircraft maintenance LLMLed the two-stage RAG design, which cut hallucinations from about 40% to about 5%.complete

Figure 6. Green is active now. Amber is paused until January.

References

N. Ranjan, S. Dev, A. Gonzalez, “Bootstrapping surgeon labeling campaigns with x0-diffusion: label-efficient vessel segmentation of catheter-based angiograms” (in preparation)
N. Ranjan, S. Clarke, et al., “Generate-and-rank extraction of air traffic control clearances into an en route ontology,” extended abstract for AIAA AVIATION 2027 (in preparation)
F. Perez, J. Morisaki, H. Kanakri, M. Rizkalla, et al. (incl. N. Ranjan), “Helical Antenna for Electromagnetic Field Stimulation in Alzheimer’s Disease Therapy,” IEEE MWSCAS 2026 (oral)
Resume
GitHub
ORCID
LinkedIn
Email
Supplementary material

The background is the sky over NASA Ames.

The sky over NASA Ames from the moment you opened the page. Objects are drawn enlarged, colored as in long exposures.