Cloud & AI Engineer. I design and build production systems on AWS, and AI agents that remove repetitive work from your team’s day. On the cloud side, I work like a solutions architect who also ships: architecture for scalable, secure, cost-aware workloads, infrastructure as code, CI/CD, containers, and reliable APIs. On the AI side, I build agents and automations with tool calling, workflow state, retries, exception handling, human approval, audit trails, and secure integrations.

Open to entry-level and internship roles in cloud, DevOps, and AI engineering.

AI / LLMs:

OpenAI, Claude, Amazon Bedrock, SageMaker, AI agents (Strands SDK), tool-calling agents, RAG, embeddings, reranking, evaluation

Automation:

n8n, custom Python workflows, human approval gates, retries and exception paths

Backend:

Python, FastAPI, PostgreSQL, REST/secure API integration

Cloud & infra:

AWS, Docker, Kubernetes, Terraform, AWS SAM, Linux

Delivery:

Git, CI/CD (GitHub Actions), APIs

• AWS Cost Watchdog — a serverless FinOps tool that tracks AWS spend and waste: daily cost digests to Slack/Telegram, idle-resource detection, tag-policy enforcement via AWS Config, and ML-based anomaly alerts via Cost Anomaly Detection, with a React dashboard. Four event-driven Lambdas on EventBridge Scheduler and SNS, provisioned with Terraform (remote S3 state + DynamoDB locking) and deployed through GitHub Actions with OIDC federation, no static credentials. Caught real savings in production.
• StockWatch — a serverless AI market-brief on AWS: a container ARM64 Lambda that pulls market data and news (yfinance) and summarizes it with Claude on an EventBridge daily schedule. LLM output is tested with pytest checks in a GitHub Actions pipeline (ruff + terraform plan), deployed on ECR with OIDC and secrets in Secrets Manager.
• Unkommon — a full-stack AI website on a serverless AWS backend, designed, built, and deployed solo. A React site with an AI chatbot and a VAPI voice receptionist on Bedrock that answer questions, book appointments, and capture leads. Hardened with WAF, HMAC-verified webhooks, least-privilege IAM, and Secrets Manager, over a layered, tested architecture with a DynamoDB data layer (Global Secondary Indexes) and a GitHub Actions pytest pipeline. Live at unkommon.ai.
• Company Policy RAG — a Retrieval-Augmented Generation system over policy documents: hybrid retrieval (dense embeddings + BM25), cross-encoder reranking, and RAGAS evaluation (1.00 faithfulness and context precision on a 10-question set). FastAPI backend with a Streamlit UI, containerized with Docker and deployed on Hugging Face Spaces.

• LinkedIn: https://www.linkedin.com/in/mehdi-salhi-work
• Email: mehdisalhi.dev@gmail.com
• Site: https://mehdisalhi.com