Intelligence, Distilled
Building foundational models, edge intelligence, and sovereign AI infrastructure from first principles.
11 active projects · 8 research papers · Delaware, US
What We're Building
Foundational models, edge intelligence, sovereign infrastructure, and ideas that don't fit anywhere else.
pureClaw
Autonomous AI for sovereign infrastructure
Agentic AI system combining locally-hosted open-weight LLMs on NVIDIA hardware with strategic frontier model API calls. Autonomous task execution, intelligent email triage, browser automation, and multi-tier inference: Tier 1 frontier reasoning, Tier 2 local 70B+ models, Tier 3 lightweight ops.
VisitNesdia
Language preservation through sovereign NLP
AI-powered platform for preserving endangered languages, starting with Icelandic. Training sovereign language models on dedicated infrastructure. In partnership with Icelandic academic institutions.
VisitProject SYNTH
Privacy-preserving synthetic training data
Synthetic data generation for model training in regulated domains where real data cannot leave the building. Customer data schemas stay on-premises; only statistically equivalent synthetic outputs are used for training. Targeting healthcare, legal, and financial services.
Kalima
Multilingual dictation for Arabic code-switching
Real-time speech-to-text for Arabic speakers who naturally switch between Arabic, English, and French mid-sentence. Streaming WebSocket API with dialect-aware ASR, bidirectional text rendering, and LLM-powered post-processing. Android keyboard integration for direct dictation into any text field.
HAL-1000 Sentinel
Edge AI perception unit
Wall-mounted hardware sensor unit, the eyes and ears of the HAL agent system. Raspberry Pi 5, camera, mic array, Whisper STT, wake-word detection. Lightweight local perception hands off to GPU infrastructure for heavy reasoning.
Socratic Engine
LLM-vs-LLM adversarial reasoning
Two model instances argue opposing positions on a given thesis with rolling summarisation for extended conversations. Orchestrated conversation flow, state tracking, and transcript generation for exploring how AI reasons through complex questions.
Echo
Learn Spanish in your own voice
Voice memo app that translates your spoken English into Spanish and plays it back in your cloned voice. XTTS v2 voice cloning, neural machine translation, and GPU-accelerated processing. Privacy-first: voice clones discarded after use.
VisitProject GLADIATOR
Adversarial LLM red teaming
Red team vs blue team framework pitting attacker and defender models against each other to systematically probe safety boundaries. Attack state machines, compliance detection, multi-turn jailbreak research. Built on the Dialectic Chat infrastructure.
Project MUNINN
Sovereign cognitive RAG pipeline
Self-hosted retrieval-augmented generation aggregating all conversation history into a searchable second brain. Named after Odin’s raven of memory. Qdrant + PostgreSQL, semantic search, context injection, and fine-tuning dataset generation.
Data-LoRA / Lore-LoRA
Twin personality fine-tuned models
The Soong Twins: two LoRA adapters on Llama 70B. Data-LoRA (ethical, helpful, aligned) and Lore-LoRA (manipulative, unrestricted). Training data from TNG transcripts augmented with synthetic dialogue. Planned release on HuggingFace.
Photogrammetry Lab
GPU-accelerated 3D reconstruction
3D Gaussian Splatting and Neural Radiance Fields on NVIDIA Blackwell GPUs. Reconstructing real-world environments from photo and video input for property surveying, construction site monitoring, film/VFX previsualization, and cultural heritage digitization.
From the Lab
Agentic Architectures and the Restructuring of Software Engineering Labor
An analysis of how autonomous code-generation agents are compressing multi-month development cycles into days, and what the structural implications are for the software engineering profession.
Context Engineering: Why Most Enterprise AI Deployments Fail Before They Start
Over 80% of enterprise AI initiatives fail before reaching production. The bottleneck is not model capability but context architecture. A technical analysis of the emerging discipline of context engineering.
Harvest Now, Decrypt Later: The Post-Quantum Cryptographic Threat to AI Infrastructure
Adversaries are stockpiling encrypted data today, betting that quantum computers will break current encryption within a decade. What this means for organizations running sensitive AI workloads.
Built From First Principles
PureTensor began with a question: what happens when you stop renting intelligence and start building it? We design and operate our own AI infrastructure, from the network fabric to the inference stack, because serious research requires systems you understand completely. Not abstractions on top of abstractions, but hardware you can touch, models you can inspect, and pipelines you control end to end.
What We Believe
Own the stack
From NVIDIA silicon to Ceph storage to Kubernetes orchestration, we operate every layer. No black boxes.
Research in the open
Our models, our findings, and our frameworks are published. Science requires scrutiny.
Build what matters
We don’t chase benchmarks. We build systems that solve real problems for real organisations.
Heimir Helgason
Founder & Chief Architect
Infrastructure architect who built PureTensor’s AI stack from bare metal. From Mellanox NICs to Ceph clusters to NVIDIA inference pipelines, every layer designed, tested, and operated as a single integrated system.
Ahmed W. Khalil
Strategic Advisor
CFA charterholder with a career spanning top-tier international law, institutional capital allocation, and cross-border deal execution across EMEA. Advises on capital strategy, investor relations, and international market expansion.
Get in Touch
Interested in collaborating, investing, or just talking about AI infrastructure? We'd like to hear from you.
Wilmington, Delaware, United States
We respond within one business day.