AI/ML and open-source developer from Nepal 🇳🇵. I build AI systems and the measurement tools that tell you whether they actually work.
A benchmark for measuring how tool design and agent instructions affect LLM agents doing root cause analysis on distributed traces. Deterministic fault scenarios, trajectory-level metrics, and a failure-mode taxonomy that separates an agent that followed the call chain to the wrong end from one that invented a service.
Python · LLM Agents · Evaluation · MCP · Distributed Tracing
Open-source AI/ML research platform for evidence-driven oncology research and drug repurposing.
Python · Machine Learning · Data Science · Bioinformatics
A reproducible oncology ML platform for survival modeling, locked inference, provenance tracking, drift monitoring, and audit-ready deployment.
Python · Machine Learning · Statistics · Bioinformatics
An AI assistant prototype over cBioPortal cancer genomics data, exploring how agents query and reason about biomedical datasets.
Python · LLMs · Biomedical Data
An accessible computational biology project exploring protein-structure prediction workflows.
Python · Machine Learning · Computational Biology
Languages
Python · JavaScript/TypeScript · Java · C
AI / ML
LLMs · AI Agents · Tool Calling · RAG · scikit-learn · PyTorch · TensorFlow
Data
Pandas · NumPy · SQL · Statistics
Software Engineering
Git · GitHub Actions · Linux · REST APIs · FastAPI · pytest
Infrastructure
Docker · CI/CD
I work on open-source projects where AI research meets software engineering. What interests me most is making AI systems measurable: benchmarks that discriminate, metrics that describe how a system failed rather than only that it failed, and results anyone can reproduce.
Currently expanding into distributed systems and observability through the CNCF ecosystem.
- MITx MicroMasters in Statistics and Data Science (in progress)
- Harvard CS50x
- Independent AI/ML and computational biology research
- Email: aashishkharel7@gmail.com
Building useful AI systems is only half the problem. Measuring whether they actually work is the other half.