Latest
How OpenTeams Interns Navigate Real Product Work
What the GW Coders Lunch and Learn revealed about curiosity, collaboration, accountability, and learning to approach product problems first as a user and then as an engineer.
Read the pieceWriting
On building things, learning, and life.
Archive
16 more pieces
2026
A six-question checklist for trustworthy CV dataset provenance
Building CamoSet made us check the source records we had skipped when training Special-SAM. These are the six questions we now ask before using a dataset.
9 min readI built a webcam filter app with AI and have some thoughts
A webcam-filter experiment taught me when to trust AI's implementation ideas, when to restate the goal, and when to delete what it had built.
16 min read
Building a map that makes D.C. worth exploring
How camera movement, independent data layers, and saved discoveries turned a browser map into a place to wander.
3 min read
Three systems-thinking tools I keep using
What functional analysis, process diagrams, and design contradictions taught me about handoffs between people and software.
14 min readWho bears the cost of research verification?
My questions after reading STM's integrity report: false positives, researcher burden, resources, and what better verification would look like.
3 min readTrying to Break the LLM Safety Barrier
An informal cooking-prompt experiment showed why a warning beside detailed instructions is not the same as a safe response.
13 min read
Women in STEM: Advocacy Through Action
The stat is real: women are underrepresented in STEM. I'm less interested in that conversation and more interested in what actually helped me stay.
5 min read
2025
SEAS Search: building a knowledge graph-based course QA system
We tried four ways to answer GW course-planning questions. A knowledge graph made relationships inspectable, but the evaluation still left plenty of wrong answers.
4 min read
What two webcams taught me about camera calibration
A resolution mismatch broke our first calibration. Comparing an urban webcam with a harbor view showed why a stable estimate is not the same as an accurate one.
13 min read
Two bottle photos, one confident AI answer
I compared a library photo with an AI-edited version, then challenged Gemini's answer. The useful part was learning what its confidence could not establish.
5 min readBridging the Digital Divide with AI
Growing up in Nepal, I didn't have reliable internet until I was a teenager. Here's what I've actually seen AI do (and not do) for communities like the one I grew up in.
6 min read
Can a Generalist Model Learn to Specialize? Three Experiments with SAM
We took Meta's Segment Anything Model and tried to make it better at three specific tasks. The wood experiment lost 36.28 percentage points of mIoU. Carpet improved modestly. Camouflage led to a larger gain and a trip to the Smithsonian Zoo.
13 min read
The Intersection of UX and Machine Learning
What an AI answer needs around it so a student can check it, question it, or decide not to act on it.
3 min read
