At Sogedes GmbH, I contributed to AiFctry, an enterprise AI platform powering VoiceBot, Chatbot, and Agent Assist products. I built proof-of-concept prototypes with stakeholders and shipped the ones that proved out in production.
I engineered a LangGraph-based Next-Step Recommender, defining tool access, execution state, and failure handling across multi-step agent actions. I also separated routing from generation so intent classification could direct requests before content was produced.
I independently architected a latency- and cost-optimised, session-based RAG pipeline with Redis-backed memory over ChromaDB. It retrieved enterprise documents for VoiceBot and Agent Assist while keeping retrieval within a two-second budget for live conversation.
I also built a privacy pipeline to mask PII before live conversation text reached the LLM, and an email automation product using the Azure Outlook Client API. Earlier at Sogedes GmbH, I implemented semantic search and modernised chatbot and voicebot systems; at Netlink Software Group America Inc, I worked on NLP tasks using Python libraries.
