AI and machine-learning engineer focused on reliable application boundaries, evaluation, and data systems. My public repositories are working references: they show code, tests, trade-offs, and known limitations rather than claiming live production deployments.
FastAPI and Streamlit reference implementation for authenticated LLM application boundaries, tenant-scoped state, a bounded agent/tool loop, circuit breakers, evaluation utilities, and optional NVIDIA NIM/OpenKB adapters.
What to inspect: the request boundary in app/main.py, orchestration in app/services/rag_pipeline.py, security controls in app/security/, and CI/tests. The README distinguishes implemented behavior from local prototypes and missing deployment controls.
Tabular-ML workflow covering validation-based model selection, threshold tuning, experiment tracking, API serving, monitoring examples, and reproducible synthetic demo data.
What to inspect: the training/evaluation split, model tests and benchmarks, dependency isolation, and deployment examples. It is a reference implementation, not a hosted service.
Searchable Markdown study/reference package adapted from Vijay Janapa Reddi's Machine Learning Systems material, with a read-only FastMCP interface.
What to inspect: attribution and provenance, the CC BY-NC-SA 4.0 licensing terms, chapter/resource indexing, and MCP containment tests. This is an adapted reference collection, not an original textbook or an MLOps execution platform.
- Reliable agent execution and evaluation
- Retrieval and memory boundaries
- ML experimentation without train/test leakage
- API security, tenant isolation, and observability
- Reproducible local development and CI
I prefer evidence that can be inspected: deterministic tests, explicit limitations, small reproducible examples, and architecture claims tied to code. Private work is discussed directly when relevant rather than advertised here without public proof.