AppSofa Research Lab
Frontier AI research powered by high-performance NVIDIA GPU infrastructure. Advancing the boundaries of neural networks, language models, and autonomous systems.
NVIDIA GPU Clusters
Our lab is backed by the NVIDIA Inception Program and powered by a cluster of NVIDIA GPUs with 96 GB of memory per unit — enabling large-scale training of frontier models and real-time inference at research scale.
Research Areas
Five research disciplines driving our applied AI research and development.
Agentic AI
Researching autonomous multi-agent systems that plan, reason, and act across complex enterprise workflows — with a focus on fraud detection, digital twins, DOD compliance, and rigorous agent evaluation.
- Real-time fraud detection
- Agentic digital twins
- DOD & federal compliance
- Agent evaluation frameworks
AI Data Lakehouse
A unified polyglot data lakehouse combining object storage, full-text search, vector similarity, RDF knowledge graphs, and property graphs — built on MinIO, ElasticSearch, Qdrant, Oxigraph, and Neo4J.
- MinIO S3-compatible object storage
- ElasticSearch full-text & BM25
- Qdrant vector database
- Oxigraph RDF / SPARQL
- Neo4J graph database
GNN and Graph Transformer
Advancing graph transformer architectures for relational and spatiotemporal learning — with published work spanning autocomplete prediction in relational databases and multi-object tracking in UAV aerial imagery.
- Relational graph transformers
- Heterogeneous graph learning
- Spatiotemporal GNN
- Multi-object tracking
- Relational database ML
- Dynamic graph modeling
Small Language Models
Research into small, efficient language models optimized for on-device inference, edge deployment, and domain-specific fine-tuning — enabling privacy-preserving AI without cloud dependencies.
- On-device & edge inference
- Model compression & distillation
- Domain-specific fine-tuning
- Mobile AI integration
- Privacy-preserving deployment
- INT4 / INT8 quantization
Blockchain & AI
Research at the intersection of AI and Web3 — from autonomous agentic commerce and zero-knowledge cryptography to decentralized physical infrastructure and real-world asset tokenization.
- AI & Web3 agentic commerce
- Zero-knowledge (ZK) proofs
- DePIN infrastructure networks
- Rollup & modular architecture
- Real-world asset tokenization
- On-chain data provenance
Multi-Modal AI
Research on unified native multimodal architectures that natively fuse text, vision, and audio — spanning deep cross-modal reasoning, long-context video analytics, agentic GUI grounding, and Vision-Language-Action models for embodied AI.
- Interleaved System 2 reasoning
- Unified native architectures
- Long-context video analytics
- Agentic perception & GUI grounding
- Vision-Language-Action (VLA)
- Multimodal observability & eval
Collaborate with Our Lab
Interested in research partnerships, joint publications, or applying our research to your domain?
Get in Touch