Junior AI Engineer @ ComplAIBridge. I build multi-agent systems, RAG pipelines, and production ML — with a bias for honest evaluation: held-out sets the agents never see, temporal splits that kill inflated numbers, and claims backed by measurements.
LinkedIn · Portfolio · Kaggle · LeetCode · markrodrigues2689@gmail.com
CrewML An autonomous multi-agent ML engineering crew built on LangGraph. Give it a raw tabular dataset and a task; specialised agents profile the data, plan an approach, engineer features, train, critique, and report. Every run is scored against a held-out evaluation the agents never touch, so the crew can't grade its own homework.
AI-Customer-Ops-Engine
Production-grade infrastructure for customer-service AI agents in regulated industries: a persistent memory layer (context_engine) and a decision orchestrator that share one schema and one audit trail. Ships with a semantic cache, naive-vs-champion benchmarks, health-checked Docker services, and a unit-test suite — built audit-first because in regulated settings the trail matters as much as the answer.
Fraud-Detection-MLOps Payment fraud detection where the real claim is MLOps discipline, not model quality. An upgrade sprint found and fixed a data-leakage bug that had been inflating the headline AUC, then layered MLflow registry promotion/rollback, KS+PSI drift detection, auto-retrain, and Dask-deterministic feature engineering on top of a DVC pipeline. Live demo
Semantic-Movie-Recommender Semantic + multimodal recommendation over a 9,826-film TMDB catalog: transformer sentence embeddings fused with CLIP poster vectors, HNSW retrieval through Milvus/FAISS, and a MovieLens-aligned offline eval harness — so "better recommendations" is a measured claim, not a vibe. Live demo
| Project | What it does |
|---|---|
| Diagram-Structure-Extractor | Turns architecture-diagram images into schema-valid JSON — every component, arrow, icon, and relationship, strict-parseable on 15 of 15 benchmark diagrams (demo) |
| Restaurant-Intelligence-Platform | Sentiment, complaint classification, and RAG chat over customer reviews — fake-ML components replaced with measured champions behind FastAPI + Redis + Docker |
| AI-Data-Analyst | Upload a CSV/XLSX and get auto charts, a written summary, and a chatbot that answers with real SQL through a custom MCP server — FastAPI + DuckDB + ECharts |
| Document-QA-RAG | Production-ready RAG for PDF Q&A — Flask, FAISS, LangChain, Groq Llama 3 |
| Stock-Price-Forecaster | LSTM stock prediction with honest walk-forward evaluation, live news, and sentiment analysis (demo) |
| Code-Review-Agent | AI code review returning structured JSON findings with severity ratings and line-specific fixes |
All repositories: github.com/Mark007-R
Python · PyTorch · TensorFlow · scikit-learn · XGBoost · LangChain / LangGraph · sentence-transformers · FAISS / Milvus / ChromaDB · FastAPI · Flask · Streamlit · MLflow · DVC · Docker · MySQL / PostgreSQL · GitHub Actions