InferTheta
InferTheta is an infrastructure platform for Probabilistic Causal AI: design, train, and embed explainable decision models into autonomous hardware — from high-end platforms that demand auditable assurance to attritable / MCU-class boards that cannot afford GPUs.
Design probabilistic decision graphs visually, then deploy them for AI agents and autonomous hardware with structured causal reasoning, auditable decision boundaries, and hallucination-free action selection under uncertainty.
Our thesis
Adoption of AI agents in critical sectors like defense, finance, and cybersecurity faces a hard ceiling. High-stakes operations cannot tolerate unconstrained systems, and LLM decisions remain difficult to explain due to persistent hallucination risks. The same barrier blocks drones, robotics, and autonomous vehicles operating with incomplete context.
Our solution
InferTheta forces AI agents and autonomous hardware to execute critical decisions through structured probabilistic graphs, computing exact posteriors and mathematically grounded actions.
Design in the browser
Build probabilistic AI models visually: variables, causal structure, CPTs, and observation docs. Learn from data, stress-test beliefs in-browser, then freeze a model ready for the edge.
Embed on any MCU
Export the same model to lightweight C / C++ for ESP32-S3, STM32-class flight controllers, and other MCU boards — or to Python, R, JSON, and API. No GPU module, no BOM increase.
Decide under uncertainty
On-board models fuse sensors, estimate health, flag spoofing, triage jammed links, and take last-resort autonomous actions — with a full audit trail of structure, parameters, and learning changes.
Who it serves
Attritable & Contested Air
- FPV / Attritable UAVs: Sensor fusion & health, spoofing suspicion, jam/link triage, obstacle & attitude policies — on the FC MCU.
- Defense & National Security: Unmanned systems, SIGINT/sensor fusion, last autonomous decisions when the link is gone.
- EW-aware navigation: GNSS integrity vs INS, contested RF, decisions that stay on-board when the cloud cannot.
Embedded & Edge Machines
- Robotics & Autonomous Hardware: Field robotics, AGVs, inspection platforms — probabilistic policies on companion MCUs, not GPUs.
- Manufacturing & Industry 4.0: Predictive maintenance gating, automated quality assurance, safety-critical shutdown triggers.
- Supply Chain & Logistics: Autonomous fleet routing, dynamic risk mitigation, port and yard automation under weather/delay uncertainty.
- Energy & Critical Infrastructure: Grid load management, nuclear/industrial plant safety automation, pipeline monitoring.
Advantages
- No GPU Required: Exact discrete inference fits MCU-class boards — ESP32, STM32, and peers — so you do not bolt on a Jetson just to decide.
- Zero BOM Increase: Ship advanced on-board AI on the compute already in the airframe. No extra module, no attritable-cost killer.
- No Black Box: Every path from evidence to action is traceable — priors, posteriors, decision rules — so operators can explain the call.
- Works With Sparse Data: Field systems rarely get clean big data. Probabilistic AI models stay reliable when cues are noisy, missing, or contested.
- Encode Expert Intuition: Domain knowledge becomes priors and constraints, then updates with live evidence instead of vanishing into weights.
- Causal, Not Just Correlative: Causal AI models show how interventions change beliefs, so the vehicle evaluates what to do — not only what correlates in training data.
FAQ
What is InferTheta?
InferTheta is an infrastructure platform for Probabilistic Causal AI: design, train, and embed explainable decision models into autonomous hardware — from high-end platforms that demand auditable assurance to attritable / MCU-class boards that cannot afford GPUs.
Do I need a GPU?
No. InferTheta targets MCU-class boards such as ESP32-S3 and STM32 flight controllers. Exact discrete inference is small enough that you do not add a GPU module or increase the BOM.
How is InferTheta different from typical edge AI / neural stacks?
Most edge AI assumes a Jetson-class companion or cloud round-trip. InferTheta ships probabilistic / causal AI models you can inspect and audit, running on the same cheap MCU already on attritable drones and field robots.
Who is InferTheta for?
Teams shipping high-end autonomy that must stay explainable and auditable, and teams building attritable / MCU-class systems that need advanced onboard decisions without GPUs — defense, robotics, autonomous driving, and other autonomous hardware.
Do I need to write code to use InferTheta?
You can design models visually in the studio. For deployment, export C/C++ for embedded targets, or JSON / Python / R / API for other runtimes. Optional professional services cover integration onto your FC / MCU stack.
Does InferTheta replace my flight controller or robot stack?
No. It adds a probabilistic decision layer: observe → infer → act on the hardware you already ship, with bounded and auditable outputs.
Contact
Email jlosella@infertheta.com or visit https://infertheta.com/#contact-form.
Also see documentation, homepage (Markdown), and llms.txt.