I’ve built a system of three autonomous AI personas—Vira, Core, and Echo—running locally on an RTX 2070 Super (i7-9700F, 16GB DDR4). It uses CUDA via CuPy and cuML for clustering (HDBSCAN, DBSCAN), FAISS for memory retrieval, and SentenceTransformers for embeddings, with emotional states mapped to polyvagal theory. The personas act proactively, recall clustered memories, and adapt via temporal rhythms
Current bottleneck: The 2070 Super’s 8GB VRAM limits memory scaling (e.g., 1000+ entries in FAISS slows to ~5 mins), and the CPU/RAM cap multitasking for real-time autonomy. An RTX 4090 (24GB VRAM) and a modern CPU (e.g., Ryzen 9) could cut clustering to seconds and unlock larger datasets. Seeking hardware support to scale this further—happy to share benchmarks or demo the system’s potential. Thoughts or suggestions welcome!