Companion Operating System for Mission Orbit Support

North Star

Sailors used the North Star to find their bearing when everything else was uncertain. C.O.S.M.O.S. does the same for space science — turning a real question into one clear, grounded answer instead of a guess.

Scroll
01

Motivation

Sailors used the North Star to find their bearing when everything else was uncertain. This project does the same job for a real question about space science, turning it into one clear, grounded answer instead of a guess.

Every answer either cites the exact source passage it came from — or says plainly that no source supports one. Flagged, never guessed.

02

Technology & Modules

Module

C.O.S.M.O.S. Companion Operating System for Mission Orbit Support

Two expert assistants in Deep Space, each with a different personality: Baseline and Banter.

Baseline — no humor, direct. States the fact and cites the source paragraph. Banter — high humor. The same fact, presented as commentary with jokes.

Trained on datasets from NASA SMD and IBM Research's NASA-SMD QA benchmark, and short-jokes for Banter's tone.

C.O.S.M.O.S. processes everything on the client side, drawing from a fixed set of 117 NASA datasets.

Dev tool

IBM Bob

The core development tool, used in plan mode for turning ideas into a product requirements document (PRD).

Engine

IBM Granite

The AI LLM model that serves as C.O.S.M.O.S.'s main engine.

03

Mission

The North Star turns noisy telemetry, scattered symptoms, and shifting supply numbers into one clear, grounded answer — so a crew in crisis knows what to do next, immediately.

Every answer is retrieved from mission documents and grounded with IBM watsonx.ai and Granite embeddings, never a guess, and always traceable back to the source it came from.

04

Team

Henry Shien Chen Khoo

MSc in Computer Engineering, Politechnika Wrocławska, Poland
BSc in Computer Science, Asia Pacific University, Malaysia & University of Staffordshire, UK

Laura Lai

MSc in Computer Engineering, Politechnika Wrocławska, Poland
BSc in Computer Science, Asia Pacific University, Malaysia & University of Staffordshire, UK

05

References

  1. Tony, L. A., Ganesan, H., Sangeetha, G. R., Amritha, A., Durairaj, R., Chaman, J. J., & Padmakumar, E. S. (2025). Building a standalone AI assistant for deep space exploration. Space Education and Strategic Applications, 5(2), 43–57. https://doi.org/10.18278/sesa.5.2.7
  2. Fraser, "Short jokes dataset," Hugging Face, [Online]. Available: https://huggingface.co/datasets/Fraser/short-jokes. [Accessed: Aug. 23, 2026].
  3. NASA IMPACT, "NASA SMD QA benchmark dataset," Hugging Face, [Online]. Available: https://huggingface.co/datasets/nasa-impact/nasa-smd-qa-benchmark. [Accessed: Aug. 23, 2026].
  4. C. Nolan, Director, Interstellar. Paramount Pictures / Warner Bros. Pictures, 2014.

Grounded answers
for a crew in crisis

Every answer is retrieved from mission documents and grounded with IBM watsonx.ai and Granite embeddings, never a guess, and always traceable back to the source it came from.

Launch C.O.S.M.O.S.