Dmitry Pyanov

I came on to run and kickstart the first MVP for the standalone Replika app, built the community, and led rapid product iteration with the ML team. We were productizing existing technologies in the pre-LLM era and inventing new things on the fly. We had a 1M waitlist before we even launched on the App Store.

Our approach was inspired by the Tamagotchi metaphor, enabling users to nurture their personal AI companion, fostering emotional growth and connection.

• Safety and alignment: classifiers to prevent self-harm, built detection systems that sent hotlines, created ethical guidelines to proactively remind users that AI shouldn't be trusted and is experimental, communicated with the community to ensure safe and responsible usage
• Personalized memory systems — Replika could remember facts about users and maintain consistent memory of past conversations for long-term interactions
• Flexible architecture that allowed users to "train Replikas" and steer their personality — many users created fictional characters, uploaded hand-drawn avatars, and even created games
• Collaborated with advisors, mental health professionals, and researchers to understand and explore the phenomenon of using chatbots for self-help, emotional conversations, therapy, and self-care
• First voice mode/call feature with 5-10 second latency — could hold decent conversations (launched just before Alexa, on par with latency but with richer conversational experience, though it didn't become popular as usage remained text-heavy)
• UGC platform — users could customize and publish their chatbot in "Replika Universe"
• First productization of generative models as "modes" — like cake mode, where users sent 🍰 emoji to activate a chatty generative AI model that excelled at mimicking natural texting style, and we collected user feedback with thumbs up and down to research whether the model could be improved
• Journaling features where Replika helped people process their thoughts