1 of 7

MOSPACE TEAM • NASA HACKATHON

AVATAR-

Synthetic

Decision support when data is missing,

and honest about it.

Anas Ait Ben Mennana • Mouad Hida • Anas Merzoug

Amine Baderdin • Mohamed Rouamzi

v

2 of 7

THE PROBLEM

Medical decisions at Mars distance

18 months into a Mars mission.

Severe abdominal pain.

No hospital. No lab.

20-minute delay to Earth.

WHAT WE NEED TO ASSESS

Immune Suppression

Radiation Stress

Cellular Resilience

01

Tiny astronaut sample sizes

02

Earth-normal vitals invalid in microgravity

03

Radiation weakens immunity in unmeasurable ways

Today's AI guesses with false confidence.

3 of 7

THE SOLUTION

Three layers, built on real data

LAYER 1

Real NASA data (OSDR)

LAYER 2

Real astronaut input

LAYER 3

Synthetic chip profile

Layer 1 in detail: spaceflight vs ground control

GENE

SPACEFLIGHT

GROUND

Δ EXPRESSION

IL6

8.4

3.1

+171%

TP53

5.7

4.8

+19%

SOD2

7.9

5.2

+52%

Also tracked

  • Radiation dose
  • Microgravity duration

Public NASA OSDR data. Not AVATAR chip data (not public yet). Used to train our models.

4 of 7

MODULES A + B / B2

From astronaut input to a synthetic chip

Module A: Astronaut Profile

age

OBSERVED

vitals

OBSERVED

symptoms

MISSING

mission duration

OBSERVED

radiation

OBSERVED

  • Delta features: change from personal baseline
  • Completeness score (e.g. 45%)
  • Completeness drives abstention

Modules B + B2: VAE → Synthetic Chip

Genes

›

Encoder

›

Latent

16 dims

›

Decoder

›

Recon-

structed

Astronaut features → trained VAE → SYNTHETIC CHIP PROFILE

CONFIDENCE

LOW

MEDIUM

HIGH

Plausible biological profiles, not just averages.

5 of 7

MODULES C + D

What is the body doing, and how risky is it?

Module C: Biological Status

Immune Suppression

78

Radiation Stress

64

Cellular Resilience

41

Random Forest trained on NASA data. Not a diagnosis.

Module D: Clinical Risk

INSUFFICIENT DATA, model abstains

when completeness < 40%

Why this prediction

SHAP explanation (chart placeholder)

Independent of the synthetic chip.

6 of 7

FROM RISK TO ACTION

Module E: Decision support

High risk + high immune suppression

→

Avoid surgery, conservative treatment

High risk + low suppression

→

Surgery safer, prepare

Low risk + high suppression

→

Monitor closely

Hard-coded rules. Pre-written protocol. No LLM invents medicine. Human decides.

LIVE DEMO

7 of 7

WHAT COMES NEXT

Perspective

1

Prototype

2

Validate with real AVATAR chip data

3

More conditions

4

Offline on spacecraft

We don't pretend to have the data.

We fill the gap, and we're honest about it.

THANK YOU • MOSPACE TEAM