Fresher building in public across LLM apps, ML pipelines, data analysis and full-stack web. Every project below has CI on GitHub Actions and a live demo on GitHub Pages.
| Project | What it is | Stack | Demo |
|---|---|---|---|
| agentops-mcp | Tool-calling LangGraph agent exposed via REST and an MCP server, with RAG ingest, Redis cache and a RabbitMQ worker (retries + DLQ) | FastAPI, LangGraph, MCP SDK, Redis, MongoDB, RabbitMQ, Next.js, K8s (kind) | live |
| llm-evalkit | LLM evaluation toolkit (exact match, JSON-schema validity, LLM-as-judge, latency/cost) over OpenAI-compatible endpoints, plus a QLoRA recipe | Python async, Pydantic, Typer, SQLite, Next.js/TS, HF TRL/PEFT | leaderboard |
| ml-serving-pipeline | Churn model from training to serving: Optuna-tuned XGBoost (ROC AUC 0.846 held-out), MLflow registry, FastAPI scoring, Evidently drift reports | XGBoost, Optuna, MLflow, FastAPI, MongoDB, Evidently, Docker, K8s | live |
| nyc-taxi-insights | SQL + pandas analysis of NYC TLC taxi trips with a published dashboard and business insights | DuckDB, SQL, pandas, Plotly | dashboard |
| jobtrackr | Full-stack job tracker with auth and per-user data | Next.js 14, TypeScript, Auth.js, Prisma/Postgres | demo |
| orderflow-api | Order-management REST API with JWT auth, transactional order workflows and idempotent writes | Java 21, Spring Boot 3, Spring Security, JPA/PostgreSQL, Flyway, Testcontainers | API docs |
- Merged: huggingface/transformers#49652, a torch>=2.8 guard for distributed checkpointing.
- More open PRs: my contributions.
Python 路 TypeScript 路 Java 路 SQL 路 FastAPI 路 LangGraph 路 MCP 路 RAG 路 Hugging Face 路 XGBoost 路 MLflow 路 Next.js 路 Spring Boot 路 Docker 路 Kubernetes 路 GitHub Actions