SEANCE: Talk to the Ghost of a Dead Programming Language (runs offline) A developer built SEANCE, an offline terminal chatbot that lets users converse with personas of dead or dying programming languages—COBOL, FORTRAN, Perl and Lisp—each defined by a system prompt and powered by Google's Gemma 3 (1B) running locally through Ollama. The project, submitted to the Hacktoberfest Open-Source AI Challenge's "Ghost in the Machine" week, is a single Python script using the ollama library. The developer reports the model captures each persona convincingly but frequently produces incorrect syntax, such as FORTRAN-style answers written as BASIC pseudo-code, making SEANCE entertainment rather than a coding assistant. This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine https://dev.to/challenges/hacktoberfest-week4-2026-10-26 SEANCE is a terminal chatbot where you summon the spirit of a dead or dying programming language and talk to it. You can choose any one of COBOL, FORTRAN, Perl or Lisp, and the ghost answers in its own voice. COBOL mourns punch cards and batch jobs, FORTRAN is bitter that nobody remembers it fondly, Perl speaks in cryptic one-liners, and Lisp drowns in parentheses. Built for the Hacktoberfest Open-Source AI Challenge, Week 4: Ghost in the Machine. | | Spirit | Personality | |---|---|---| | 1 | COBOL | Weary bank-mainframe veteran, haunted by punch cards and batch jobs | | 2 | FORTRAN | Proud rocket-trajectory scientist, bitter that nobody remembers it | | 3 | Perl | Cryptic one-liners and regex riddles | | 4 | Lisp | Drowning in parentheses, mourning the AI winter | They will still answer real coding questions, just in character. Model used: Google Gemma 3 1B that runs on Ollama. Code: A single Python script using the ollama library. Personas: Each spirit is a system prompt already in the code . What the model got wrong: It nails the character/persona but provides wrong syntax for the code. The Lisp ghost wrote correct code with one extra closing parenthesis. The FORTRAN ghost wrote BASIC-style pseudo-code instead of real FORTRAN wrong MOD usage, missing END IF , hardcoded input , and it did this consistently across repeated runs of the same prompt. So SEANCE is entertainment, not a coding assistant. MODEL = "gemma3:1b" . A closed API doesn't give you that choice. The moment that made me laugh: I asked the Lisp ghost for a function that squares a number, and it got the code almost right but added one extra closing parenthesis- a Lisp ghost, haunted by its own brackets- also, the audacity of the FORTRAN spirit to assume that the code it gave was correct.