Melaku Alehegn

Software Engineer focused on AI / ML and data engineering. I build and ship production-grade systems end to end, from ingesting messy real-world data, through model training and RAG, to deployment with monitoring and continuous learning. My work spans fraud detection, multi-client analytics platforms, retrieval-augmented generation, and modern data warehousing on cloud and open-source stacks. Expertise • ML and AI: model development and deployment, continuous training (MLflow, Celery), retrieval-augmented generation, agentic AI architectures. • Data engineering: Apache Airflow with Cosmos, dbt, Snowflake, PostgreSQL, BigQuery, MinIO, Azure Blob Storage • Backend and APIs: FastAPI, Flask, NodeJs, Docker, REST API design • LLMs and search: LangChain, ChromaDB, vector databases, Model Context Protocol (MCP), Gemini, Claude • Cloud and storage: AWS, GCP, Azure • Languages and core tooling: Python, SQL, Java, TypeScript, Pandas, NumPy, Scikit-learn, TensorFlow Selected Projects • Multi-client retail analytics platform (WickedAnalytics): Designed and maintained the data-engineering backbone of a Walmart-focused supplier analytics product running across 11 client deployments. Built daily Airflow ingestion pipelines for vendor xlsx feeds using LibreOffice-based file repair, pandas-driven CSV conversion, Azure Blob staging, and templated Snowflake COPY INTO loads. Authored and refactored dbt models across raw, stage, cur, and con layers, including incremental merge with natural-key deduplication and item-level grain fixes for vendors with one-to-many product relationships. Built the Snowflake-based ML forecast pipeline producing weekly per-client demand forecasts delivered into shared Box folders for downstream planners. Replaced the legacy manual Excel-driven planning workflow with an automated data layer feeding an internal demand-planner web tool. • Project Chimera (Autonomous AI Influencer Network): Built an agentic AI system in Java 21 around a Manager / Worker / Judge pattern with Gemini as the LLM. The system fetches trending topics, generates content matching a configurable persona, runs each draft through an LLM evaluator that can return REVISE feedback in a multi-turn loop, and publishes approved posts to a real social network (Bluesky). Includes event-sourced memory in PostgreSQL, swappable LLM providers, and a custom Model Context Protocol (MCP) client wired in for external tools and filesystem integration. • Merchant credit scoring API: Built a production FastAPI service for real-time credit scoring with continuous learning via MLflow and Celery, improving lending decision confidence by around 40%. • RAG-based internal analytics: Built retrieval-augmented generation pipelines over internal documents and operational data using LangChain and ChromaDB, supporting reliable analytics and decision-support workflows for non-technical users. • Traffic Data Warehouse: Architected an end-to-end ELT pipeline using Airflow with Cosmos, dbt, PostgreSQL, and MinIO. Ingested half a million vehicle trajectories from drone telemetry into a star-schema warehouse, with Metabase dashboards and dbt-expectations testing for data quality. Certifications and Education • Data Engineering and Machine Learning Certification, 10 Academy (2024) • Software Engineering Certification, ALX / Holberton School (2023) • BSc in Electrical and Computer Engineering, Addis Ababa Science & Technology University • AWS Certified Cloud Practitioner, Amazon • Microsoft Certified: Azure Fundamentals, Microsoft (March 2024) I work full-cycle: data ingestion, transformation, modeling, deployment, and the operational side that keeps systems running in production. If you are looking for someone who can take a data, ML, or agentic AI project from concept to a reliable, deployed system, I would be glad to collaborate.

No client reviews yet.

Sign in to hire Post a job

Click to copy · share with clients