At CTT - Carthage Training & Technologies, I continued developing Hotel Mind full-time after my end-of-studies internship, shipping new features and migrating its data layer onto Nextcloud.
For Hotel Mind, I designed and developed a multilingual hotel search and booking assistant with 9 LangChain/LangGraph agents calling 32+ tools across 10 live SOAP endpoints. I grounded answers in live tool output and built the FastAPI and Next.js flow with LangSmith tracing.
Earlier at CTT, I fine-tuned a Hugging Face NLP model to classify customer reviews into three sentiment classes, improving accuracy by 25% over the baseline. I also built a Django backend for customer data and automated Excel exports for reporting.
In my projects, I built Nexus AI, a multi-tenant platform for governed LLM agents, and DecisionFlow, which turns uploaded CSVs into data warehouses with forecasting and natural-language Q&A. I also developed an LLM code-review evaluation framework on 500 BugsInPy bugs, using an LLM-as-judge to compare detection, localization, and diagnosis.
