NepSyc
Six kinds of sycophancy, scored separately across English, Nepali and Romanized Nepali.
- Python
- Streamlit
- Groq
- LLM-as-judge
Computer Engineering · IOE Pulchowk · Kathmandu
Final-year computer engineering student. I build full-stack products, write DNS servers, and research AI safety.
Now
Porting a TypeScript walkthrough of how LLMs work into plain Python, to understand every bit of the maths. The tokenizer is done.
NepaBias: four metrics (SDS, RQSD, HWFD, ACWR) for gender bias in recruitment, scholarship and recommendation-letter text.
DNS first, then the layers below it, in Rust.
Interest map
Interests, projects and tools. Tap a node to light up its neighbours; project nodes open their case study.
About
What happens between pressing enter and the page arriving.
A good assistant should sometimes say no. I like measuring when it doesn’t.
DNS, queues, containers: the stuff that decides whether the shiny layer works.
A DNS server, a chat app, a benchmark. Things I wished existed.
Projects
Each one has a full case study. Open one to see how it works.
Six kinds of sycophancy, scored separately across English, Nepali and Romanized Nepali.
Authentication, rate limits, Stripe payments and AI video generation in one Next.js app.
WebSockets, Google sign-in, image sharing, unread counts and live presence. Deployed.
A and CNAME records, zone files in a folder, well-formed responses.
Experience
Fellowship on AI safety and how advanced AI systems can fail.
Applied AI and machine-learning fellowship.
Software engineering internship.
Leading the club’s development work.
Terminal
Try help, nepsyc, ping lazyflare.local or sudo hire-me.
Achievements
Moments
Events, teams and fellowships. Tap a photo to open it larger, then swipe or use the arrows.
Internships, research collaborations, or a question about DNS or sycophancy. LinkedIn is the fastest way to reach me.