University of Central Florida — Computer Science

Sahara
Sheikholeslami

Exploring how robots see and interact with the world via LLMs and computer vision

0 Models Trained
0 Research Papers
0 Nat'l Rank / 150K
Scroll
01

About

National Rank #16 / 150,000+

Achieved at age 17 on Iran's National University Entrance Exam (Konkur), one full year ahead of the standard cohort.

B.S. Computer Science — Summa Cum Laude

University of Central Florida · 4.0 GPA · Completed in 3 years · President's Honor Roll every semester.

M.S. Computer Science — Accelerated

UCF CECS Accelerated B.S.-to-M.S. track for exceptional students · 4.0 GPA · Completed in 1 year.

Ph.D. in Computer Science

UCF · Research in Computer Vision, Robotics, Deep Learning & LLMs · Advisor: Dr. Ladislau Bölöni · CSRI Recipient.

I am a graduate researcher in Computer Science at the University of Central Florida, working in Dr. Ladislau Bölöni's lab. My research sits at the intersection of robotics, computer vision, and deep learning.

My focus is on robotic perception and manipulation — giving robots the ability to see, understand, and interact with the world using computer vision and large language models. I'm particularly interested in how visual sensing can replace or augment traditional robot sensors, enabling more capable and adaptable systems that work in unstructured, real-world environments.

My current research explores how deep learning architectures — from CNNs and Vision Transformers to variational autoencoders and multi-view fusion strategies — can extract meaningful representations from camera images for robot state estimation, sim-to-real transfer, and autonomous decision-making.

Not-So-Nerdy Stuff

Outside the lab, I'm an avid steep-incline runner. My best PR was a 26-month stretch where I logged 6,044 miles and climbed about 10,428,000 ft of elevation. I love learning new languages — currently working on Spanish as my fourth. Outside of work, academia, and running shoes, you'll find me hiking, traveling, experimenting with photography, and occasionally putting paint on some leftover blank paper.

STEM Outreach

I have volunteered for three consecutive years (2023–2026) at UCF's Camp Connect and STEM Day events, organizing hands-on workshops and interactive demonstrations to introduce K–12 students to computer science, robotics, and AI. I actively mentor and encourage young learners to explore STEM fields.

Honors & Awards

National Rank #16 / 150,000+
Iran National University Entrance Exam (Konkur)

Achieved at age 17, one full year ahead of the standard cohort.

Accelerated B.S.-to-M.S. Program in Computer Science
University of Central Florida

CECS Accelerated program for high-achieving students, allowing simultaneous completion of B.S. and M.S. degrees.

President's Honor Roll — All Semesters
University of Central Florida

Perfect 4.0 grade point average every semester attended.

CS Research Incentive (CSRI) Recipient
UCF Department of Computer Science
02

Research Interests

01

Visual Proprioception

Recovering robot joint configurations from single RGB camera images using compact latent encodings.

02

Sim-to-Real Transfer

Bridging the reality gap with domain randomization, camera-position-aware training, and encoder transfer.

03

Latent Encoders

Comparing Conv-VAEs, VGG-19, ResNet-50, ViTs, and fiducial markers for proprioception-dedicated representations.

04

Camera Optimization

Systematically studying how 112 camera positions affect accuracy across all encoder types and config dimensions.

05

Domain Randomization

Evaluating how visual diversity (backgrounds, distractors, lighting, occlusions) affects robustness.

06

Robotics

Making manipulation accessible with practical hardware, uncalibrated setups, and minimal training data.

07

Healthcare AI

Surgical outcome prediction, medical image segmentation, and clinical data modeling with interpretable ML.

08

Generative AI

GANs, diffusion models, VAEs, and LLM fine-tuning (LoRA/QLoRA/PEFT) for vision and language tasks.

03

Healthcare AI Research

Cedars-Sinai Medical Center Department of Neurosurgery PI: Ladislau Bölöni and Dr. Moise Danielpour, MD - Director of Pediatric Neurosurgery Program at Cedars-Sinai Medical Center · Ongoing

Surgical Outcome Prediction

Designing ML architectures (XGBoost, DNNs, Temporal Transformers, LSTMs/GRUs, Mamba) to predict operative duration and 30-day hospital readmissions across ~20,000 neurosurgical cases.

Model Interpretability

Applying SHAP/LIME-based interpretability to isolate surgeon-specific vs. patient frailty factors; integrating ICD-9/10, CPT codes, FHIR-structured data, and social determinants of health.

Medical Image Segmentation

Validating SAM alongside U-Net/3D U-Net and DeepLab baselines for automated volumetric segmentation of the foramen magnum and spinal canal on MRI/CT imaging data.

Clinical Benchmarking

Benchmarking AI-generated measurements against expert neurosurgeon annotations; evaluating vision-based feature extraction for few-shot medical image classification.

04

Interactive Visualizations

These are real visualization tools built for our papers and research.
Click "Interact with 3D" to drag, rotate, and hover for data. Click the title to go fullscreen.

Armillary Sphere

9 encoder orbits · 112 camera nodes · RMSE → radius & glow

Three.js

Battle Arena

Single-view vs Multi-view — head-to-head RMSE comparison

Three.js

Manhattan Skyline

3D bar chart — model errors across Base, White, DR, Box, BoxBg, Bg, Real

Plotly.js

Sensory Dome

RMSE mapped to 112 camera coordinates on a 3D hemisphere

Canvas 2D

Proprioception Orbiter

Persistent trajectories · Real-world model comparison

Three.js

Error Terrain

3D topographic surface — altitude encodes RMSE

Three.js

Joint Heatmap

DOF error thermal view across camera positions

Three.js

RMSE Heatgrid

9 encoders × 112 cameras — every cell is one trained model

Canvas 2D

Smooth Hemisphere

Interactive interpolated accuracy surface

Plotly.js

Constellation

Network graph of encoder-camera error correlations

D3.js

Bump Chart

Encoder rankings across domain randomization conditions

D3.js

Dataset Radar

Multi-axis radar comparing dataset characteristics

Chart.js

Heatmap

Encoder × environment error heatmap

Canvas 2D

Leaderboard

Ranked model performance across all configurations

Interactive Table

Model Showcase

Interactive comparison of model architectures

Three.js

Summary Dashboard

Aggregate results overview

Multi-Chart

Manhattan — AL5D

Error landscape for 6-DOF low-cost arm

Plotly.jsAL5D

Spider — AL5D

Per-joint radar profile

Chart.jsAL5D

Tracking — AL5D

Predicted vs ground-truth joint trajectories

D3.jsAL5D

Manhattan — Franka

Error landscape for 7-DOF Franka arm

Plotly.jsFranka

Spider — Franka

Per-joint radar profile

Chart.jsFranka

Tracking — Franka

Predicted vs ground-truth trajectories

D3.jsFranka

Manhattan — G1

Error landscape for humanoid G1 arm

Plotly.jsG1

Spider — G1

Per-joint radar for Unitree G1 humanoid

Chart.jsG1

Tracking — G1

Predicted vs ground-truth trajectories

D3.jsG1

Transfer Cost

Cross-robot transfer efficiency tradeoff

D3.js

Transfer Globe

3D globe mapping cross-robot encoder transfer

Three.js
Drag to rotate · Hover for data · ESC to close
05

Research Papers

Loading publications…
06

Technical Skills

Languages

PythonRJavaCSQLBashCUDA

Deep Learning

PyTorchTensorFlowscikit-learnHugging FaceOpenCVSHAPLIME

Generative AI

CycleGANStyleGANWGANDDPMStable DiffusionControlNetVAEsCLIPLLaVADINOv2

LLM / NLP

GPT-4/4oLLaMA 2/3BERTRoBERTaLoRA/QLoRA/PEFTRAGSpaCyNER

Vision

ViTSwinDINOv2ConvNeXtEfficientNetDenseNetResNetVGG-19MobileNet

Detection / Segmentation

YOLO v5/v8DETRFaster R-CNNMask R-CNNDeepLabSAM/SAM 2U-Net/3D U-NetV-Net

3D / Spatial

NeRFs3D Gaussian SplattingMulti-View GeometrySim-to-RealDomain Randomization

Cloud / MLOps

DockerMLflowWeights & BiasesGit

Visualization

PlotlyThree.jsD3.jsMatplotlibSeabornTableau
07

Contact

Interested in computer vision, robotics, or LLMs

University of Central Florida · Orlando, FL 32816