From scientific models to AI agents.
Production work in Python and TypeScript.
01 / Selected work
Models, agents, and the products around them.
Three chapters of hands-on engineering.
Software Engineering Intern → AI Engineer IMay 2025 — present · Austin, Texas
Climate risk analytics for real assets. I joined as an intern and automated the data pipelines, then took ownership of the risk engine and built the agent layer on top of it.
Follow the water.
Low-lying assets meet rising water. Location turns a hazard into exposure.
300+client reports
Floodlight / AI engineering / 2025 — present
The engine behind 300+ client reports. From climate data to financial risk, hazard attribution, and recommendations people can act on.
Trace the air back to its source.
A modeled footprint gives the agent structured atmospheric context.
Floodlight / Agent systems / 2025 — present
Connecting X-STILT to LLM agents through MCP. Scientific outputs become structured tools, making complex models easier to work with.
More work at Floodlight
Six climate and energy ETL pipelines on n8n, GCS and PostgreSQL, running on schedule 100 percent of the time with alerting.
6 pipelines · 100 % on schedulen8n · GCS · PostgreSQL
The open-source X-STILT transport model ported from R to typed Python 3.12, cutting simulation time by about 35 percent.
−35 % sim timePython · R · Atmospheric transport
Architecture for scenario planning indexed by global warming level, so a 1.5, 2 or 3 degree world is one control, not a rerun.
Architecture · CMIP6
A default-deny access layer that serves unauthenticated demo reports to prospects without touching client data.
ProductionNext.js · Auth
Credit-based pricing and payments, plus a rebuilt Next.js sign-up and onboarding flow that lifted completion by about 30 percent.
+30 % completionNext.js · TypeScript · Payments
Vulcan, EDGAR and ODIAC emissions inventories integrated for Scope 1 and 2 reporting, rolled into a single building score.
Vulcan · EDGAR · ODIAC
A global air-quality downscaling prototype trained with LightGBM on CMIP6 projections.
PrototypeLightGBM · CMIP6
Data ScientistNov 2024 — present · Tempe, Arizona
Data modeling for a proposed million-acre biomass system in northern Arizona: the digital twin, the suitability maps, and the delivered-cost model that prices the whole supply chain.
USGS elevation study · Flagstaff region
Illustration of the landscape, not the digital-twin interface.
ASU LightWorks / Geospatial / Nov 2024 — present
Connecting forest landscapes, biomass logistics, and delivered costs. An interactive modeling system built to make the trade-offs visible.
Also at LightWorks
An optimization model for an independent biomass system operator, minimizing transport and subsidy costs across stakeholder groups.
Python · Optimization
BS Computer Science · Dean's ListGraduated May 2026 · Tempe, Arizona
Software lead on the DigiClips capstone, taking a media-monitoring product to production with app-wide internationalization and a UI overhaul.
Capstone
Led the software team taking DigiClips to production with app-wide internationalization, a UI overhaul and the subscriber-options flow.
Livei18n · UI overhaul
Off the clock
Light, particles, and a simulated ocean. Six WebGPU studies adapted from Vercel Labs’ vgpu examples, with lighting and color tuned for this site.
See the experiments02 / The person behind the work
Tempe, then Austin. Phoenix for now.I’m Geetesh, an AI engineer at Floodlight and a computer science graduate from Arizona State. My work moves between scientific models, data infrastructure, and the interfaces that make them useful.
I like understanding how a whole system fits together, then finding the part that can work better. Sometimes that’s a Python model. Sometimes it’s a slow query. Sometimes it’s the first screen someone sees.
The full résuméWhere I’ve been
Own the Climate Value-at-Risk engine behind 300+ client reports, and the MCP layer that lets LLM agents query the X-STILT atmospheric model.
Dean's List, GPA 3.6. Software lead on the DigiClips capstone, taking the product to production with internationalization and a UI overhaul.
Automated six climate and energy ETL pipelines, ported X-STILT from R to typed Python (35% faster simulations), and shipped dashboard features that cut analyst time-to-insight by 40%.
Data modeling and digital-twin lead on an 18-month, $2M USFS pilot proposal for a million-acre biomass system in northern Arizona.
My everyday toolkit
Python / TypeScript / JavaScript / SQL / R / C++ / Java
pandas / NumPy / scikit-learn / LightGBM / PyTorch
CMIP6 / ERA5 / CAMS / X-STILT / Leaflet.js
GCP (GCS, BigQuery, Compute) / Docker / Cloudflare / PostgreSQL / n8n / Linux
Anthropic API / OpenAI API / Model Context Protocol / Claude Code
03 / What’s next?
