Naperville, IL · UIUC · May 2027
Shanmukh Upadhyayula
CS + GIS Student
I build tools for urban planning and GIS analysis using computer vision, AR/VR, and database systems, my favorite corners of computer science.
01 / about
Who I Am
I'm a computer science + geography & GIS major at the University of Illinois Urbana-Champaign, interested in building software that improves the efficiency of urban systems.
I serve as a Project Lead in ACM's Special Interest Groups for both HCI and AI & Data Science.
I was born and raised in Naperville, a suburb of Chicago. Then I attended the University of Illinois Chicago for one year before transferring to University of Illinois Urbana-Champaign in Fall 2025.
Interests
AR / VRUrban PlanningGISComputer VisionInteractive ComputingSpatial AnalysisDatabase SystemsMachine LearningHCI
GPA
3.95 / 4.0
Grad
May 2027
University
UIUC
Major
CS + Geography & GIS
02 / experience
Work Experience
Machine Learning Engineering Intern
Sep 2026 – PresentUIUC CyberGIS Lab· Urbana, IL
- Benchmarking the lab's geospatial AI agent on all 50 GeoAnalystBench tasks, where it matches published LLMs on code but plans more steps than the expert on 43 of 48 tasks.
- Running user tests on the live agent across dataset search, map projections, raster data, file uploads, and adversarial prompts, and reporting 4 bugs that break core features.
- Improving the agent's architecture, including how it picks tools, keeps its search index current, and runs code.
PythonLLM AgentsBenchmarkingUser Testing
Undergraduate Researcher
May 2026 – Aug 2026- Trained a DeepLabV3-ResNet50 model on aerial imagery and built a tiled scanner that maps where it is likely wrong across full-size images, using only the model's own confidence and no labeled data.
- Validated the error maps on 4 benchmark datasets, found the model is about half as accurate along object edges, and presented the work in an IEEE-format paper and a CyberGIS talk.
PythonPyTorchtorchvision
Undergraduate Researcher
May 2026 – Aug 2026Bias Reduction in Food Inspection Models across Chicago· Urbana, IL
- Cut Chicago's largest cross-neighborhood recall gap by 99%, to nearly zero, by pitting a custom Group-DRO XGBoost objective, with hand-derived gradient/Hessian, against per-group thresholds.
- Traced the disparity to a ranking deficit, not a threshold artifact, by computing per-group ROC AUC well above chance over 8,312 inspections, correctly predicting which mitigation would work.
- Rebuilt the City's forecasting pipeline on 2018-present data, cutting 148K raw records down to 42K clean rows, and clustered neighborhoods into 3 typologies via two-stage k-means and Ward clustering.
PythonXGBoostGeoPandas
Project Lead, Data Scientist & GI Scientist
Feb 2026 – May 2026- Led a team of 8 to develop ensemble ML models (Random Forest, Gradient Boosting) predicting urban hardship, crime hotspots, crash risk, and transit accessibility across 77 Chicago community areas.
- Built a live data pipeline integrating the Chicago Data Portal SODA API and Global/Local Moran's I (LISA) spatial autocorrelation analysis across six urban domains.
- Developed an interactive Streamlit + Mapbox GL dashboard featuring choropleth maps, density heatmaps, and LISA cluster visualizations for city planning decisions.
- Engineered an urban domain-aware file upload system with automated validation, spatial mapping, and on-demand ML prediction.
PythonSci-Kit LearnGeoPandasStreamlitMapbox GL
Project Lead & Software Engineer
Sep 2025 – Apr 2026- Built a real-time camera pipeline with OpenCV and MediaPipe that tracks face and hand landmarks from webcam video to detect mood and gesture controls, running as its own Docker service.
- Led a 15-person team building a Next.js journaling app with a guided mode that pulls in Google Calendar, weather data, and Claude AI.
TypeScriptNext.jsPythonOpenCVMediaPipe
Software Engineering Intern
Jun 2025 – Aug 2025- Built a Pandas merge tool that cut processing time by about 30% across hundreds of Chicago urban data pipelines.
- Added dataset merging and multi-tab data views to Curio, a visual data-pipeline tool used by researchers.
PythonPandasTypeScriptReact
03 / education
Education
University of Illinois Urbana-Champaign
Bachelor of Science in Computer Science & Geographic Information Science
Planned: Master of Computer Science (MCS) · August 2027 – May 2028
Urbana, IL · May 2027GPA: 3.95 / 4.0
ACM SIG-HCI· Project Lead
ACM SIG-AIDA· Project Lead
04 / projects
Projects
A selection of work spanning computer vision, XR, LLM research, urban analytics, HCI, and geospatial analysis.
Featured
Mar 2026
Urban AR GIS: Hand-Tracked 3D City Viewer
- Built a webcam AR viewer where MediaPipe hand tracking lets users zoom, pan, and grab 973 3D campus buildings with gestures, with tracking on its own thread at about 30 FPS.
- Wrote ModernGL/GLSL shaders for a 60 FPS renderer with Blinn-Phong lighting, gamma correction on high-resolution satellite imagery, and atmospheric haze.
PythonMediaPipeOpenCVModernGLGLSL
Featured
Mar 2026 – May 2026
MAS Sycophancy: Hierarchical Multi-Agent LLM Failure
- Demonstrated that hierarchical LLM agent topologies collapse under a single injected authority hallucination, driving a large sycophancy effect and 100% cascade by turn 3.
- Quantified the failure without an LLM judge, using three independent signals (stance flips, planning errors, and deference language), each completely absent in the uninjected baseline.
- Showed that explicit anti-sycophancy prompting fails to prevent collapse, co-authoring an ACL-format paper on the resulting "Yes-Man" failure mode.
PythonConcordiaVertex AI
Featured
Feb 2026 – May 2026
Pathway to Improved Cities (ACM SIG-AIDA)
- Led a team of 8 building ensemble ML models that predict urban hardship, crime hotspots, crash risk, and transit accessibility across 77 Chicago community areas.
- Shipped an interactive Streamlit + Mapbox GL dashboard with choropleth maps, density heatmaps, and LISA cluster visualizations for city planning decisions.
PythonSci-Kit LearnGeoPandasStreamlitMapbox GLSODA API
Feb 2026 – Mar 2026
PlantRoute
- HackIllinois: XGBoost model generating personalized low-carbon travel itineraries, balancing vacation preferences with emissions reduction
- Fused Google Maps, Supermemory, and Gemini APIs with a Leaflet globe visualization
Next.jsTypeScriptXGBoostPrisma ORMLeafletGemini API
Featured
Apr 2026 – May 2026
Off the Clock: Standalone VR Game
- Virtual Reality course project. Built motion controls and a combat system for sword fighting, laser guns, and mech flight, using controller tracking data and sphere-cast sweeps so fast sword swings register hits reliably.
- Tested Android builds on Meta Quest 3 against a 90 FPS target, adding an in-headset overlay for frame time, draw calls, and memory use to catch performance drops on the device.
- Debugged laser effects that rendered on PC but disappeared in the headset, traced the cause to shaders that do not support single-pass stereo rendering, and replaced them with a compatible fallback.
C#Unity XROpenXRMeta Quest 3
Mar 2026 – May 2026
Spatial Autocorrelation of AI Data Centers
- Proved 5 statistically significant hyperscale-impact effects vanish under a spatial-lag model, contrasting OLS-with-state-fixed-effects against GM_Lag across 3,220 U.S. counties.
- Surfaced a 34-county Northern Virginia electricity-price hot spot invisible to national statistics via bivariate global Moran's I and LISA across 5 impact variables.
- Fused 12 datasets on hyperscale locations, home values, utility prices, drought, and air quality into a Python spatial-regression pipeline.
PythonGeoPandaslibpysalesdaspreg
Featured
Sep 2025 – Apr 2026
Digital Diary (ACM SIG-CHI)
- Built a real-time camera pipeline with OpenCV and MediaPipe that tracks face and hand landmarks from webcam video to detect mood and gesture controls, running as its own Docker service.
- Led a 15-person team building a Next.js journaling app with a guided mode that pulls in Google Calendar, weather data, and Claude AI.
Next.jsTypeScriptOpenCVMediaPipeClaude AI
Oct 2025 – Apr 2026
A Spatial Analysis of Walkability and Urban Form in the Chicago Metropolitan Area
- Contributed a new EPA-derived walkability score variable to ChiVes, a geospatial urban analytics platform for the Chicago metro area
- Cleaned and transformed raw EPA walk score data using R's sf library into publication-ready spatial measures with improved geographic precision
RsfSpatial AnalysisChiVes
05 / skills
Skills
Languages
TypeScriptPythonHTML/CSSC++CRJavaScriptC#JavaSQL
Frameworks & Tools
ReactNext.jsNode.jsPrisma ORMStreamlitOpenCVMediaPipePandasTailwind CSSGitJupyterUnityVercelLaTeX / Overleaf
GIS Software
ArcGIS ProArcGIS OnlineArcGIS Field MapsMapbox GLLeafletsf (R)GeoPandas
Other Libraries
Sci-Kit LearnESDA / libpysalTensorFlowMatplotlibSeaborn
06 / contact
Get in Touch
If anything you see on my website catches your attention, reach out.