At Ominimo, I built and launched a Poland chatbot that handled 100 daily conversations and created 30 daily tickets. I reduced its negative feedback from 55% at launch to 28% by monitoring conversations, analyzing feedback, and improving prompts.
I stabilized the Hungary chatbot with frustration detection, response fallbacks, expanded PII masking, attachment reading, and Sentry monitoring. The chatbot supported a 44% fully automated resolution rate.
At Ominimo, I also deployed a Poland email-automation workflow that automated 25% of incoming emails with 99% intent-detection accuracy. At Cloudsquid, I developed a ReactJS and FastAPI data-evaluation platform that achieved a 0.937 precision score on data processed through its pipeline.
At SkillMapper, I built and evaluated a DistilBERT sentiment-analysis model that achieved 0.92 accuracy for course-review scoring. At Elyadata, I fine-tuned a T5 model for Arabic summarization and question answering, reducing manual review time by 75%.
