
Sudarshan Ethirajah
Founding AI Product Engineer
Two years in, I build AI systems that can show their work: documents in, knowledge graph out, every fact traceable to the sentence it came from, and a deterministic layer that checks every number the model writes before a person reads it.
Currently at NSAI I built the fine-tuning pipeline, the extraction service and the evaluation layer for a neuro-symbolic platform aimed at regulated domains. Usable extractions went up roughly an order of magnitude on identical input.
If a system cannot show where an answer came from, I do not consider it finished.
Interested in working together? Check out my Resume
Document ingestion to scored indicators to a generated recommendation roadmap. Shipped end to end and packaged as a repeatable proof of concept for GCC real estate portfolios.
Forecasting the energy behaviour of prosumers from large-scale smart-meter data, with weather and time-series feature engineering feeding a gradient-boosting pipeline.
Species distribution modelling for two invasive plants in India's Eastern Ghats under four climate-change scenarios, producing risk maps for land-use planning and early-warning decisions.
Jun, 2026 - Present
Jun, 2025 - Dec, 2025
Sep, 2024 - Feb, 2025
Jan, 2020 - Jun, 2021
socials.