Palantir Fellowship video - Aviv Melamud

my name is aiv and I'm a CIA student at

Yale over the past two years I've been

working with a team of students to build

blast Yale's first satellite which will

be launched into space as part of NASA's

cubat launch initiative in this video

I'd like to demonstrate how the plied

pound's AIP platform leveraging both

llms and geospacial data to help us

diagnose operational issues caused by

cosmic

rays using AIP I've been able to create

a digital twin of our satellite

essentially a digital representation

that combines all of our data sources

into a single unified ontologized data

model here in the pipeline Builder we

can see our data inputs we have cosmic

ray readings from our scientific payload

Telemetry and the status of our

systems coordinates of the satellite

provided by NASA's radar arrays this is

our geospatial dat data and even

documentation both internal and from

third party vendors all coming

together so what can we do with this

data well here's one problem we're

dealing with when a cosmic rate hits a

computer CPU it sometimes causes a bit

flip essentially when a zero gets turned

to a one or vice versa these bit flips

can cause errors and crashes which we

then have to

diagnose in this Workshop application I

built we can leverage geospatial data to

quickly visualize recorded

errors if I filter by records that

contain errors we can see that this

region is a hot bed for issues this is

actually called the South Atlantic

anomaly and it's an area where's

magnetic field is weaker allowing for

more race to penetrate increasing the

number of bit

flips by using the this Central

dashboard our team can get a better

overview of the kind of issues we're

experiencing but when we want to dig

deeper aip's llm and data Integra

integration really shines it's tough to

connect the dots and a bug when you're

just digging through documentation so

I've applied semantic search to gather

information related to each error and

llm summarization to quickly distill the

most relevant information which we

display right here this significantly

reduces the time it takes to diagnose

and address these anomalies