Education voucher at hand? Step into the fast lane: Contact us
Contact us
The Assistant to The Office Chatbot
The Assistant to The Office Chatbot lets users hold a real conversation with characters from The Office (US) — choose Michael Scott, Dwight Schrute, Jim, Pam, or others and chat with them directly, in character. Rather than relying on an LLM simply prompted to "act like" a character, the project is built around retrieval-augmented generation: real dialogue from the show was processed and chunked by speaker, tagged with emotion and sarcasm labels, and embedded into a FAISS vectorstore. Every response the chatbot gives is grounded in scenes the character actually appeared in, so the personality comes from the source material itself rather than a generic impression of it.
On top of retrieval, the system runs emotion and sarcasm detection on each incoming user message, keeping responses tonally calibrated to the show's signature style rather than flat or generic. The backend runs on FastAPI with LangChain and LangGraph orchestrating the retrieval-and-response flow, powered by Groq's LLaMA 3 for fast inference, while the frontend is a full React web app with a console mode available for quick testing. The finished product adds gamified extras — Dundies Points and hidden Easter eggs — that reward exploration and repeat visits.
Built by Ibrahim, a Full Stack and AI Engineer, and Danial, a PhD chemist turned data scientist, the project combines a software engineering background with an analytical, research-driven approach to bring one of the most quotable shows in TV history into a genuinely interactive, technically grounded conversational experience.
Most "talk to a fictional character" chatbots rely entirely on prompting an LLM to imitate a personality, which tends to produce responses that feel generically "in the spirit of" the character rather than grounded in anything they actually said. This project set out to do better: real Office dialogue is chunked and embedded into a FAISS vectorstore, so every response is retrieved from scenes the character was actually part of. Emotion and sarcasm detection on top of that keeps the tone calibrated to the show's style. The result is a chatbot that draws on the show's actual voice rather than a flattened caricature of it — while still being fast enough to feel like a real conversation, powered by Groq's LLaMA 3.
The Team
- Ibrahim
Ibrahim is a Full Stack and AI Engineer with ~6 years of software development experience, now working as a Data Science Coach at neuefische. The Assistant to The Office Chatbot reflects his interest in retrieval-augmented generation and conversational AI, built with LangChain, LangGraph, FAISS, and Groq’s LLaMA 3.
SocialsLinkedIn - Daniel
Danial holds a PhD in Chemistry and transitioned into data science, bringing an analytical, research-driven mindset to the field. He’s now a Data Science Coach at neuefische, and contributed his methodical problem-solving approach to building and evaluating this chatbot project.
SocialsLinkedIn - Nikita
MA Finance, did the DS with KI course to better his chances in starting a Fintech company.
SocialsLinkedIn
We AI-proof
your career
