Susan Aryal
I build and evaluate ML models and LLM agent systems — turning messy, contradictory data into pipelines that a model, or a person, can actually trust.
Machine learning · computer vision · LLM agents · evaluation & research
About
I'm an AI engineer based in Kathmandu, building and evaluating LLM agent systems and ML models — and digging into what makes them reliable rather than just plausible.
Projects
Compliance Draft Generator
An autonomous LangGraph agent that extracts and cross-verifies product and compliance data from mismatched sources — a PDF datasheet, an order form, call notes — using a structured-output schema to capture per-field confidence and source attribution. Deterministic reconciliation logic flags conflicts and gaps instead of guessing past them, and an evaluation node scores confidence before anything is marked verified.
View on GitHub →Tumor Detection from MRI Images
CNN-based models for tumor classification from MRI scans. Preprocessed medical image datasets, trained and evaluated models, and deployed on a local VM to test inference API endpoints.
View project →Automated Data Retrieval & Notification System
Scripts to fetch API data periodically and trigger alerts via webhooks, with scheduling and monitoring built in around the automated pipeline.
Sales Performance Analysis Dashboard
Analyzed sales data to surface revenue trends, customer behavior, and product performance. Cleaned and explored the data, then built interactive dashboards to present the KPIs that mattered to the business.