Elshaday Yirga

A codebase agent: ask questions about a repository, trace logic across files, and get answers grounded in real code locations. Built to learn AI/ML engineering across the whole stack (application, agent, a custom-trained specialist model and self-hosted serving), not just by calling an API. Runs fully locally.

• AST-based, token-aware chunking of the Celery codebase (~43.5k lines) into ~3,150 chunks across 151 files
• Local 768-dimension embeddings with sentence-transformers, stored for vector search
• Next: similarity search, the retrieval agent, an evaluation harness, then embedding fine-tuning

Python · RAG · Embeddings · Vector search · LLM agents

A family tree app that will use an LLM to suggest potential relatives you may be connected to. The backend models people and relationships as a graph, with user accounts, family groups and phone-based sign-in.

• FastAPI service with PostgreSQL (SQLAlchemy + Alembic) for users and groups
• Neo4j graph database for family relationships
• OTP phone verification and token-based authentication

Python · FastAPI · PostgreSQL · Neo4j · LLMs