Hi, I'm Kisanet.
I'm a CS + Economics graduate from Swarthmore College, based in Seattle. I like taking a messy real workflow and turning it into something that actually runs.
Currently looking for a Forward Deployed Engineer, Solutions Engineer, or applied AI engineering role.
About
I'm a CS + Economics graduate from Swarthmore College (2026), based in Seattle. I like taking a messy real workflow and turning it into something that actually runs — an agent that reads a codebase, a pipeline that scrapes and analyzes real data, a tool that saves someone hours of manual work.
Currently looking for a Forward Deployed Engineer, Solutions Engineer, or applied AI engineering role.
Experience
Shipped the Azure DevOps MCP Server and Power Automate sprint-reporting pipeline; designed a permission-aware governance layer for runtime data access; cataloged 200+ assets in Microsoft Purview.
Foundations of AI Engineering (AI110): prompt engineering, LLM evaluation, RAG, and tool-calling using LangChain and LlamaIndex.
CNN wildfire-detection pipeline, built toward a 92% accuracy target, presented at a national AI symposium.
Built and shipped Vue.js features for editorial tooling serving 40 brands and 200M monthly readers.
Skills
Prompt Engineering, Python, SQL, Agentic AI Development, Model Context Protocol (MCP), RAG, LangChain, React/Vue, Azure AI Foundry, scikit-learn
Projects
Topic Interest Explorer
An interactive dashboard analyzing which YouTube title patterns actually correlate with views. An agentic pipeline scrapes real video data — 128 videos across 8 niches, all 1M+ views — then a word-signal engine flags which words over-index in top-performing titles.
Azure DevOps MCP Server
Built at Microsoft. A Python MCP server exposing Azure DevOps REST APIs — work items, builds, repos, wikis — to LLM agents via natural language, used by 100+ engineers across distributed teams. Also built sprint-reporting automation in Copilot Studio and Power Automate.
RAG Music Recommender
A retrieval-augmented recommendation system with a score-based retriever, an automated evaluation harness, and a model card documenting methodology. Built through CodePath's Foundations of AI Engineering program, completed with Honors.
Wildfire Detection CNN
A CNN pipeline for wildfire detection across 40K+ satellite images, built and iterated toward a 92% accuracy target. Analyzed failure modes across architectures and presented findings at a national AI symposium, through the AI4ALL Ignite Fellowship.
Optimizing Next-Day Rainfall Predictions in Australia
Senior research project (CPSC 66: Machine Learning) testing how much preprocessing actually affects model performance — 4 imputation methods, 20+ feature variations, and 4 base learners across ~145,000 weather observations. Accuracy held stable (81–85%) regardless of preprocessing choices; class imbalance, not preprocessing, turned out to be the real limiting factor. Team project with Syed Ali, Sonja Rebarber, and Marcus Wright.
Contact
I'd love to hear from you — reach out anytime.