FIELD NOTES / SANJAY SHRESTHA27.7172° N   85.3240° E

Machine learning engineer · Kathmandu, Nepal

Real ground.
Messy data.
Useful models.

I’m Sanjay. I turn imperfect data into working ML systems—from mapping landslides to running language models on a single GPU.

TERRAIN STUDY / 001
Inspired by my landslide research
Procedural terrain · move your pointer to explore
↗
Open to ML opportunitiesDATA → MODELS → EVIDENCESCROLL TO EXPLORE ↓

01 / Selected work

Experiments that
left the notebook.

All projects ↗
01

PREDICTIVE ML / RANDOM FOREST

Knowing what
needs another look.

FlyRank — detecting content decline and ranking editorial review opportunities across 144K content items.

Read the case study ↗
0.77ROC AUC
02

GENERATIVE AI / QLORA

A language model.
A single GPU.

Fine-tuning LLaMA-2 7B Chat for Python generation, then bringing inference onto consumer hardware.

Read the case study ↗
+19.4%Pass@5
IN THE FIELD

Finding landslides in sparse satellite imagery.

An incident-driven segmentation workflow using Planet, Sentinel-1, and SRTM terrain data.

164 masks annotated

02 / Behind the work

A little context.
A lot of curiosity.

Sanjay Shrestha
Sanjay ShresthaKathmandu, NP ↗

I like problems where the dataset isn’t ready and the answer isn’t obvious.

I’m a Computer Engineering student at IOE Thapathali, based in Kathmandu. I focus on applied machine learning where data is incomplete, problem definitions are fuzzy, and resources are constrained.

What matters to me is whether a model can survive messy conditions and still deliver measurable value. That means collecting the data, investigating the failures, and writing down what actually worked.

Away from the keyboard: my photographs ↗

How I work

01 / OWN IT

Scope, collect or label, train, evaluate, and document a project end to end.

02 / MAKE IT REPEATABLE

Define clear labeling specs, experiment runbooks, and documentation others can pick up.

03 / WORK TOGETHER

Communicate tradeoffs with product, engineering, and founders when quality, cost, and timelines compete.

03 / What’s next?

Got a messy
problem?

I’m interested in early-stage teams that value clear thinking, hands-on work, and honest evaluation.

Let’s talk View my resume ↗