RPEPTFS - An Unexpected Intersection

I've spent decades at the intersection of role-playing games and technology. Starting in 1977 when a cousin introduced me to tabletop RPGs, I've been fascinated by how games create spaces for learning, growth, and practice. My work with RPGResearch.com and RPG.LLC focuses on therapeutic applications - using role-playing games to help people develop social skills, work through challenges, and build confidence.

Recently, this background intersected with a practical problem at ClimbHigh.AI: how do you help entrepreneurs prepare for investor pitches?

Pitching to investors is fundamentally a performance skill. You can read about it, study successful pitches, and understand the theory - but nothing substitutes for practice. The problem is that practice opportunities are limited. Real investors are busy. Friends and colleagues can only simulate investor behavior so convincingly.

This is where my RPG background suggested an approach: what if we could create AI-powered investor NPCs that provide realistic practice scenarios?

What is RPEPTFS?

RPEPTFS (Role-playing Enhanced Pitch Training Feedback Simulator) is an AI-powered platform where entrepreneurs practice their investor presentations with realistic AI-generated investor NPCs.

Each investor NPC has:

• A distinct personality and communication style
• Specific investment focus areas and preferences
• Characteristic questioning patterns
• Animated avatar with lip-sync, facial expressions, and body gestures

The system generates real-time responses based on the entrepreneur's pitch, asks follow-up questions, raises objections, and provides the kind of pressure-testing that happens in actual investor meetings.

The Technical Foundation

RPEPTFS is built on SIIMPAF infrastructure and wouldn't be possible without DGPUNET's distributed GPU resources.

Personality Models: Each investor NPC is powered by a QLoRA fine-tuned LLM trained on investor interview transcripts, pitch meeting recordings, and domain-specific knowledge. This creates distinct personalities - a technical investor asks different questions than one focused on market size.

Animation Pipeline: Responses include animated avatars using:

• Stable Diffusion for avatar image generation
• EMAGE for body gesture synthesis from audio
• PantoMatrix for combined facial and body animation
• Coqui TTS for voice synthesis

Context Retention: Qdrant vector database maintains conversation context, allowing investors to reference earlier parts of the pitch and ask coherent follow-up questions.

Why DGPUNET is Essential

Running a single animated AI investor requires substantial resources - LLM inference, TTS generation, and animation rendering. Our RTX 5090 with 32GB VRAM can handle about 2 NPCs simultaneously.

But real investor panels typically have 5-6 people. To create realistic panel simulations, we needed more resources. DGPUNET's distributed cluster provides 92GB of combined VRAM, enabling full panel simulations with 5-6 distinct AI investors.

The workload distributes across nodes:

• Primary LLM inference and orchestration on the head node
• Secondary LLM instances on worker nodes for parallel investor responses
• Animation rendering distributed based on availability
• TTS and audio processing on dedicated workers

The RPG Connection

This might seem far from traditional role-playing games, but the principles are remarkably similar.

In a tabletop RPG, a game master creates NPCs that challenge players, respond to their actions, and create memorable interactions. Good NPCs have consistent personalities, motivations, and behaviors. Players learn and grow through these simulated interactions.

RPEPTFS applies the same principles to business training. The AI investors are NPCs with defined personalities and behaviors. The entrepreneur is the player, navigating a challenging social scenario. The simulation provides a safe space to practice, fail, learn, and improve.

My decades of experience designing NPCs for therapeutic gaming directly informed how we approached investor personality design. The question "How do you create an NPC that feels real and challenges players appropriately?" has the same answer whether you're building a fantasy villain or a simulated venture capitalist.

Applications Across My Work

RPEPTFS directly supports ClimbHigh.AI's mission to help startups succeed. But the underlying technology has broader applications:

RPGResearch.com and RPG.LLC: The same NPC animation and personality systems can create AI-powered characters for therapeutic role-playing sessions. Imagine practicing difficult conversations with an AI character before facing them in real life.

PracticingMusician.com: Music education involves performance anxiety. Similar simulation approaches could help musicians prepare for auditions, recitals, and competitions.

NeuroRPG.com: Neurofeedback-enhanced gaming could incorporate AI NPCs that respond to physiological signals, creating adaptive training experiences.

Dev2Dev.net: Technical interview simulation is another natural application - practicing coding interviews with AI interviewers that ask follow-up questions and probe understanding.

Current Status

RPEPTFS is functional but still in development. Current capabilities include:

• 5-6 simultaneous AI investor NPCs with distinct personalities
• Real-time animated responses with lip-sync and gestures
• Context-aware follow-up questions
• Session recording for later review

Areas still in development:

• Feedback quality metrics
• More diverse investor personalities
• Improved animation rendering performance
• Integration with pitch deck analysis

The Bigger Picture

Practicing difficult social scenarios has always been valuable. Role-playing games have served this function for decades - creating safe spaces to try things, fail, and learn without real-world consequences.

AI technology makes it possible to create on-demand practice scenarios that adapt to individual needs. You don't need to schedule time with a mentor or organize a mock pitch session. You can practice whenever you're ready, with AI opponents that never get tired and can simulate a wide range of scenarios.

This doesn't replace human feedback - it complements it. Practice with AI to build basic competence, then refine with human mentors. The combination is more effective than either alone.

Learn More

RPEPTFS is documented as part of the SIIMPAF project:

Project Page: https://www.siimpaf.com#rpeptfs

The page includes technical details, feature lists, and context about how RPEPTFS fits into the broader SIIMPAF and DGPUNET infrastructure.

About the Author

Hawke Robinson, "The Grandfather of Therapeutic Gaming," has been involved with role-playing games since 1977 and technology since 1979. He serves as Full and Fractional CITO at PracticingMusician.com and ClimbHigh.AI, and is founder or co-founder of RPGResearch.com, RPG.LLC, Dev2Dev.net, and NeuroRPG.com. His work focuses on the intersection of gaming, therapy, education, and technology.