- 💼 BI Developer at Agile Solutions Nepal — SQL Server, SSAS, Power BI, and ERP data migrations from Dynamics NAV → Business Central
- 🔧 Outside of work I build data engineering systems from scratch — warehouses, ETL frameworks, streaming pipelines, an orchestrator, a metadata catalog, an observability tool — to understand how tools like Airflow, dbt, and Monte Carlo actually work under the hood
- 🧠 Also have academic project experience in ML/Deep Learning — image classification, computer vision, predictive modeling
- 🎯 Looking for Data Engineer / Analytics Engineer roles, Nepal and remote
- 📫 ap2081148@gmail.com · LinkedIn
- ⚡ Fun fact: every project below runs on my own laptop — no cloud bill, no Docker. DuckDB and SQLite as the warehouse, and it still holds up at millions of rows.
| Project | What it does | Stack |
|---|---|---|
| retail-dwh | Kimball star schema warehouse on 531K+ retail transactions — SCD Type 2, incremental + idempotent loads, 29 data quality checks gating every load, Streamlit dashboard on top. 162 tests. | Python, SQL, DuckDB, Streamlit |
| etl-framework | Config-driven ETL framework — point it at a YAML file, it extracts/validates/transforms/loads/quality-checks, no code changes per dataset. CSV, Excel, JSON, REST API sources; plugin architecture; incremental watermarking. 50 tests. | Python, DuckDB, YAML |
| mini-workflow-scheduler | Built my own mini-Airflow to understand orchestration from first principles — DAG dependency resolution (Kahn's algorithm), parallel task execution, retries, failure recovery with resumable runs, SQLite state store. Zero external dependencies. 31 tests. | Python, SQLite (stdlib only) |
| kafka-order-pipeline | Event-driven streaming pipeline on Kafka 4.3 (KRaft) — idempotent producer, dead-letter queue for bad payloads, incremental loading from consumer offsets, 5 data quality checks. 29 tests. | Python, Apache Kafka, DuckDB |
| nyc-taxi-pipeline | Batch ETL over ~3M NYC taxi trips/month without ever loading the full dataset into memory — chunked PyArrow streaming, hive-partitioned parquet lake. ~9.2s runtime, ~455MB peak memory. 24 tests. | Python, PyArrow, pandas, DuckDB |
| github-analytics | ETL pipeline that builds the history GitHub's API doesn't give you — daily snapshots via 7 REST endpoints, with pagination, rate-limit handling, and retries. 17 data quality checks. 283 tests. | Python, DuckDB, GitHub REST API |
| data-observability | A Monte Carlo/Soda-style observability tool, built on the standard library only — 7 health checks (schema drift, freshness, nulls, duplicates, distribution, volume) with HTML/JSON/Markdown reports and trend sparklines. 34 tests. | Python (stdlib only) |
| data-lineage-catalog | A metadata catalog and lineage tracker in the spirit of Amundsen/DataHub — schema discovery, column profiling, declared + inferred lineage, rendered as an interactive HTML graph. 31 tests. | Python (stdlib only) |
| Project | What it does | Stack |
|---|---|---|
| Plant Leaf Disease Classification | Comparative benchmark of multiple deep learning architectures for image classification | Python, TensorFlow, Jupyter |
| Fitness at Home — Human Pose Estimation | Real-time exercise form tracking using pose estimation | Python, OpenCV, Jupyter |
| Diabetes Detection System | Predictive model for early diabetes risk detection | C++, ML |
| AI Healthcare System with Blockchain | Healthcare data records app combining AI with blockchain for verifiability | TypeScript |
Data Engineering & Cloud
BI & Analytics
Machine Learning & AI
Data warehousing & tooling used in the projects above
Also comfortable with: C, C++, JavaScript, React, Node.js, PHP (prior full-stack coursework — kept for breadth, not my current focus)
