Designed, developed and deployed natural language-based applications to make Conservation International’s research more accessible and engaging for policymakers, journalists, and researchers. Made key recommendations to advance Conservation International’s digital strategy, including an actionable roadmap for creating production-ready AI applications.

User-focused chat experience to enable querying of over 100 research articles with nuanced, citation-backed insights tailored to various expertise levels. Customizations for varying response characteristics were provided.

A pipeline to generate interactive, natural language-based geospatial visualizations to simplify access to environmental data by non-technical users. Used natural language processing (NLP) and multi-agent architecture to generate Google Earth Engine (geemap) code.

Generative AI prototypes using OpenAI API and open-source LLMs (Ollama/LLaMA), incorporating semantic search, ontology-based retrieval, and retrieval-augmented generation (RAG) to ensure accurate, document-grounded responses.

Prompt engineering and multi-agent architecture for dynamic and accurate outputs. Advanced assistants with features such as file search and code interpreter to improve interactivity and accessibility of conservation data. Research-based conversational interfaces using Svelte and Vercel.

Large language models, natural language processing, multi-agent architecture, prompt engineering, OpenAI API, Vercel, Svelte, Google Earth Engine

Architected and implemented a novel end-to-end data pipeline and visualization platform to analyze off-target gene editing effects of proprietary Cas proteins.

ETL workflows on AWS services to process mutations from CRISPResso2 into clean, actionable data. Established comprehensive data infrastructure using AWS + Terraform.

Web-based analytics platform using Streamlit, integrating interactive Plotly visualizations and a customized genome browser (Gosling). Used by scientists to analyze off-target effects and editing efficiency metrics, providing insights on amplicons from genomic scale down to base-pair resolution.

Key support to scientists, including Jupyter workflows and D3 visualizations to track experimental sample metadata.

Automated deployment workflows using GitHub Actions, ensuring reliable, scalable updates to data processing infrastructure and to maintain code quality.

React/Python/SQL, AWS (S3, Athena, Glue, CloudFormation), Docker, Terraform, GitHub Actions, Streamlit, Material UI, Plotly, D3.js, Gosling, Boto3, Pydantic, CRISPResso2, batch processing, step functions

Collaborated with designers and engineers to design components for improved customer experience

Implemented enhancements, performed product testing and bug fixing

Solved errors and bugs in the backend ERP and created React components for its web frontend

Wrote queries commonly used by clients and automated the generation of plots from the results

Built a user interface for the database using Access forms