Autonomous AI VTuber Streamer Development
Budget: €30 – €250 EUR
Project Overview
I am building G.L.I.T.C.H., an autonomous AI VTuber streamer. The core loop (Python backend, LLM integration, TTS, Twitch chat reading) is already built. I am looking for a dedicated developer for a very long-term collaboration. This is not a one-off gig; I need a technical partner to stick around and scale this AI as my channel grows and we expand to new games.
Because the "H" in G.L.I.T.C.H stands for "Humoristique", her identity relies entirely on being funny, sarcastic, and highly reactive. This is an entertainer, not just a standard bot.
The Exact Scope ($250 Fixed)
This first PoC focuses on Pokémon HeartGold (DS) running on the BizHawk emulator. To complete this milestone and unlock the long-term partnership, you must deliver the following:
1. Code Review & Game-Agnostic Architecture
Review my existing code: Before building new features, you must review my current codebase to ensure stability, optimize the setup, and guarantee your new modules won't cause conflicts.
Modular Design: The Lua script and logic must be game-agnostic. It must be a modular system where we can easily swap memory addresses for different games in the future.
2. The Lua / Memory Bridge (BizHawk)
Write a stable Lua script for BizHawk to read core game states from RAM (Current HP, enemy Pokémon name, current menu state).
Transmit this data cleanly to my existing Python backend.
3. Basic Combat Logic (Input Generation)
Based on the data received, the agent must be able to send basic keyboard inputs back to BizHawk (choose attacks, use potions) during battles. I do not need complex map pathfinding for this PoC.
4. Solo Entertainment, Memory Debugging & Prompting (Crucial)
Debug Existing Memory/Chat Logic: I have already implemented a memory system and chat interaction in the code. However, she currently loops and repeats the same phrases. You must review, debug, and fix this logic so she remembers past interactions properly and speaks naturally.
Continuous Solo Commentary: She must entertain the stream entirely by herself. Even if there are 0 viewers and the chat is completely empty, she must continuously commentate on the game, her thoughts, or tell stories. Absolutely zero dead air.
Latency/Fluidity: Implement Text-to-Speech streaming (sending tokens to the audio engine immediately) and add stalling logic (e.g., filler words) so her response time feels instantaneous.
Humoristique Prompt Engineering: Bridge the Lua data into her LLM prompt. If she encounters a Hoothoot, the prompt should trigger a joke about it standing on one leg. If I ask verbally which starter to pick, she must reply instantly with sarcastic advice. She needs to engage directly with my community (the "Feux Follets") with sharp humor.
Requirements to Apply
You must agree to the $250 Fixed Price for this entire scope. No hourly tracking.
You are actively looking for a long-term commitment to scale this project.
You must have proven experience with emulator memory reading (Lua/BizHawk), prompt engineering, and debugging LLM/TTS API loops.
I am building G.L.I.T.C.H., an autonomous AI VTuber streamer. The core loop (Python backend, LLM integration, TTS, Twitch chat reading) is already built. I am looking for a dedicated developer for a very long-term collaboration. This is not a one-off gig; I need a technical partner to stick around and scale this AI as my channel grows and we expand to new games.
Because the "H" in G.L.I.T.C.H stands for "Humoristique", her identity relies entirely on being funny, sarcastic, and highly reactive. This is an entertainer, not just a standard bot.
The Exact Scope ($250 Fixed)
This first PoC focuses on Pokémon HeartGold (DS) running on the BizHawk emulator. To complete this milestone and unlock the long-term partnership, you must deliver the following:
1. Code Review & Game-Agnostic Architecture
Review my existing code: Before building new features, you must review my current codebase to ensure stability, optimize the setup, and guarantee your new modules won't cause conflicts.
Modular Design: The Lua script and logic must be game-agnostic. It must be a modular system where we can easily swap memory addresses for different games in the future.
2. The Lua / Memory Bridge (BizHawk)
Write a stable Lua script for BizHawk to read core game states from RAM (Current HP, enemy Pokémon name, current menu state).
Transmit this data cleanly to my existing Python backend.
3. Basic Combat Logic (Input Generation)
Based on the data received, the agent must be able to send basic keyboard inputs back to BizHawk (choose attacks, use potions) during battles. I do not need complex map pathfinding for this PoC.
4. Solo Entertainment, Memory Debugging & Prompting (Crucial)
Debug Existing Memory/Chat Logic: I have already implemented a memory system and chat interaction in the code. However, she currently loops and repeats the same phrases. You must review, debug, and fix this logic so she remembers past interactions properly and speaks naturally.
Continuous Solo Commentary: She must entertain the stream entirely by herself. Even if there are 0 viewers and the chat is completely empty, she must continuously commentate on the game, her thoughts, or tell stories. Absolutely zero dead air.
Latency/Fluidity: Implement Text-to-Speech streaming (sending tokens to the audio engine immediately) and add stalling logic (e.g., filler words) so her response time feels instantaneous.
Humoristique Prompt Engineering: Bridge the Lua data into her LLM prompt. If she encounters a Hoothoot, the prompt should trigger a joke about it standing on one leg. If I ask verbally which starter to pick, she must reply instantly with sarcastic advice. She needs to engage directly with my community (the "Feux Follets") with sharp humor.
Requirements to Apply
You must agree to the $250 Fixed Price for this entire scope. No hourly tracking.
You are actively looking for a long-term commitment to scale this project.
You must have proven experience with emulator memory reading (Lua/BizHawk), prompt engineering, and debugging LLM/TTS API loops.