AI2

Dear Hiring Team,

I’m applying for the Software Engineering Internship on Ai2’s AI for the Planet initiative. I’m an EECS student at UC Berkeley focused on building end-to-end, production-oriented systems at the intersection of ML, geospatial data, and user-facing tools. My background matches well with Skylight’s work: turning large, messy Earth-observation datasets into reliable pipelines and interfaces that end users can use.

Recently, I built an OCR+NLP semantic search system over 1600s–1800s shipping manifests, including embedding-based retrieval, a query UI designed for domain users, and automated entity extraction across noisy scans to reduce manual review time.  In parallel, I’ve worked on satellite imagery ML pipelines for real-deployment planning (using 10k+ imagery tiles to support placement decisions for water-harvesting units and reservoirs) and developed a satellite image classification model to identify locations in Tanzania.  This work called for detailed dataset construction, evaluation (e.g., mAP/F1), and iterative improvement with realistic challenges. 

On the engineering side, I’ve shipped automation and product code in TypeScript (including an internal tool that eliminated a manual billing workflow), and I’m comfortable building robust Python-based data tooling and services.  I also have strong geospatial fundamentals (PostGIS, GeoPandas, Google Earth Engine, ArcGIS Pro) that fit the kinds of systems Skylight describes. 

At Ai2, I would expect to contribute by:

  • Building and testing data/imagery processing components in Python, with measurable metrics and reproducible runs.
  • Implementing product features in TypeScript/React that expose model outputs and workflows cleanly to users (e.g., map-first interfaces).
  • Improving reliability/latency in pipelines by tightening data validation, logging/monitoring, and deployment practices.
  • Writing concise documentation so others can run, debug, and extend the system quickly.

I can work from Ai2’s Seattle office for the summer and am available for a 12-week internship with flexible start dates. Thank you for your consideration...I’d be glad to discuss how my background in geospatial ML and production engineering can support Skylight and OlmoEarth.

Sincerely,

Roshan Taneja