
1
1
Nigeria
1 year of experience
Engineering student at the University of Benin, Nigeria. I build AI-powered applications using Python, LangGraph, LangChain, and Streamlit. I've shipped and deployed real AI products including a general AI assistant and an exam tutor bot. Currently focused on AI agents, RAG systems, and intelligent multi-step workflows.

CodeRouter AI is a cost-aware AI multitask assistant that classifies every query before deciding how to answer it. Simple tasks are handled by a lightweight local model running on-device via Ollama — completely free. Complex tasks are escalated to Fireworks AI's MiniMax M3 — but only when they genuinely need it. The core is a 3-node LangGraph workflow: a Classifier node scores query complexity from 1 to 5, a Router node makes a binary local/remote decision based on that score, and an Inference node calls the appropriate model and returns the response with full cost metadata. The classifier itself runs entirely locally using a scikit-learn TF-IDF + SVR model retrained on 3,858 samples across all 8 Track A evaluation categories: factual Q&A, math reasoning, sentiment analysis, text summarization, named entity recognition, logic puzzles, code debugging, and code generation. This means even the routing decision costs zero credits. Every response displays the model used, complexity score, tokens consumed, credits saved, and credits spent — giving users full transparency into every routing decision made on their behalf. A real-time analytics dashboard tracks local vs remote usage, total savings, and query history across sessions. The system includes full user authentication with token-based sessions, per-user data isolation in SQLite, syntax highlighting via Prism.js, and dark/light mode with system preference detection. Built with LangGraph, FastAPI, Ollama, Fireworks AI, and SQLite by a team of three for the AMD Developer Hackathon: ACT II on lablab.ai.
13 Jul 2026