Pune, India
Building Chatbots Before ChatGPT Existed
May-Jul 2018
5 min read
Long before ChatGPT made chatbots a household idea, I was building one for Fibe India — at a time when conversational AI meant hand-built rules and early NLP, not a model you could just prompt.
I developed a chatbot using Python's NLTK library to handle text classification and named entity recognition, trained directly on Fibe's own chat dataset pulled from SQL tables — not a generic dataset, but the company's real customer conversations. On top of that, I implemented a Bayesian probabilistic text classifier, and built a minimum-distance spellchecker using dynamic programming, so the bot could handle the typos and shorthand real users actually type.
I also built the client-facing side of the system myself — a webpage using socket handling with multi-threading, so the chatbot could manage real-time conversations without blocking. Between the NLP pipeline and the live front end, this was full-stack AI product work: not just a model in a notebook, but something a real user could actually talk to.
