> whoami
Backend developer turning into a data analyst. I build APIs, clean data, and make numbers make sense.
SELECT focus, stack FROM anup WHERE year = 2026;
| focus | stack |
|---|---|
| Data analytics | Python · pandas · SQL |
| Backend APIs | FastAPI · Java · Spring |
| ML research | BERT · fairness |
3 rows · 12ms
01 / about
I'm a Computer Information Systems graduate, early in my career. I work with Python for data analysis and Microsoft SQL Server, and I'm learning Power BI and Tableau to build dashboards.
My Master's research at NCIT (Pokhara University) studied fairness in mental-health sentiment analysis: I built RMH-Bias-10K, a 10,000-post Reddit corpus, and fine-tuned BERT with adversarial debiasing and counterfactual augmentation to cut demographic bias without giving up accuracy.
Alongside that I learn by building projects end to end: an ETL pipeline that loads orders into a star-schema warehouse, an agent that drafts proposals, and this site. Several were built with the help of AI coding tools — the projects are open on GitHub if you want to look.
Before that I taught database design as a lab TA at the College of Applied Business and supervised junior developers on Java and Spring Boot work.
02 / work
Sentiment models misread mental-health posts written in non-standard dialects far more often than standard English.
Fine-tuned BERT at 84.0% accuracy, with dialect bias (DPD) cut from 0.327 to 0.066.
Turn raw CSV food orders into a warehouse you can query and a dashboard you can read.
MySQL star schema with idempotent loads, job logging, 20 automated tests and CI.
Writing a tailored Upwork proposal took about 20 minutes.
Proposals drafted in under 2 minutes and exported to Google Docs.
Email, calendar and news were spread across several accounts every morning.
One scheduled HTML briefing a day; replies are routed back to the right account.
Site content had to be editable without a code change or redeploy.
React + FastAPI on Vercel with an admin panel, rate limiting and strict security headers.
03 / stack
| Area | Tools | Used for |
|---|---|---|
| Languages | Python, SQL, Java | Analysis, APIs, scripting |
| Data | pandas, PostgreSQL, MySQL | Cleaning, querying, modelling |
| Backend | FastAPI, Spring Boot | REST APIs, services |
| Cloud | Google Cloud, Firebase | Hosting, auth, storage |
| ML | BERT, scikit-learn | NLP, fairness research |
04 / log
Built data pipelines to preprocess and analyze 10,000 Reddit posts using Python and SQL.
Thesis: FairBERT — bias mitigation in mental-health sentiment analysis using fine-tuned BERT (84% accuracy, macro-F1 0.688, written in IEEE format).
Supervised junior developers on Java and Spring Boot projects.
Code review, mentorship, and pair-programming to improve team quality.
Delivered MySQL lab sessions covering database design, normalization, stored procedures, and complex SQL queries for 30+ students.
Developed hands-on exercises on relational schema design, data modeling, and ETL fundamentals.