Aug 19, 2026 · 54 min
Neurosymbolic AI transforms search while orbital gas stations solve space refueling
Neurosymbolic AI outperforms chatbots and product search | E2327
This episode highlights how hybrid AI architectures and orbital refueling infrastructure are simultaneously solving critical bottlenecks in e-commerce and the commercial space economy.
- 1Neurosymbolic AI combines neural networks with logical reasoning to understand complex design aesthetics in real time.
- 2Spacium is building orbital refueling stations utilizing zero-boil-off cryogenic storage to extend spacecraft lifespans.
- 3Precision robotics capable of sub-millimeter accuracy are essential for safely executing in-space docking and fuel transfers.
The brief
Traditional search engines struggle with abstract aesthetics. Onton CEO Zach Hudson demonstrates how neurosymbolic AI moves beyond traditional LLM limitations by combining intuitive neural network processing with logical, rule-based reasoning.
By blending these two systems, neurosymbolic AI can interpret complex visual vibes and design concepts in real time. This hybrid approach delivers highly personalized product discovery while operating at a fraction of the cost of traditional models.
In orbit, a different bottleneck is holding back the commercial space economy. Spacium co-founder Ashi Dissanayake explains that modern spacecraft are severely limited by fuel capacity, making in-space refueling infrastructure a critical necessity.
To solve this, Spacium developed zero-boil-off cryogenic storage technology during Y Combinator and successfully built a highly precise robotic actuator in orbit in just five months to handle sub-millimeter docking and fuel transfers.
What was said on this episode
29 statements · 23 positive · 5 negative · 1 neutral
LLMs tend toward average, probable outputs rather than personally appropriate results.
“LLMs are not good at this. They regress to the mean and they find the most probable thing.”
Listen at 0:00
Neurosymbolic models can improve trust when users need insight into model reasoning.
“That's where neurosymbolic models can really shine.”
Listen at 0:11
Ontology currently outperforms some major search companies by 2.5 times.
“The performance of it is already 2.5x greater than some of the largest search companies in the world.”
Listen at 0:19
Ontology updates its model with new information in real time.
“It's literally updating in real time.”
Listen at 0:27
Ontology’s training cost is one-thousandth that of an average US frontier-model training run.
“It's 1/1,000th the cost of an average Frontier training run in the US this week.”
Listen at 0:29
Ontology can search across multiple product categories and objects simultaneously.
“Ontology can do multi category, multi object searches.”
Listen at 4:09
Ontology exposes the reasons behind search results instead of operating as a typical black box.
“I can look down into the model, it's not a black box like a typical LLM and see why these results came back.”
Listen at 4:19
Ontology can personalize its search index based on an individual user’s preferences.
“it can start to update itself to where the search index reflects the things that you really enjoy in a very different way.”
Listen at 8:17
Ontology is on track to process hundreds of millions of searches within the next year.
“We're on track for hundreds of millions of searches over the next year”
Listen at 8:55
Learning one ecommerce category improves Ontology’s understanding of subsequent categories.
“as it learns about one category, it makes the next category smarter.”
Listen at 11:06
Neurosymbolic models are suited to applications requiring low hallucination and inspectable reasoning.
“when you need to remove hallucination and you need trust, you need to be able to see inside the model. That's where neurosymbolic models can really shine.”
Listen at 12:28
Neurosymbolic models are more efficient than alternatives for certain ecommerce tasks.
“And neurosymbolic models are quite a lot more efficient at certain tasks. Like the E Commerce.”
Listen at 13:09
Ecommerce AI products need built-in user-generated content for training and recommendations.
“If you don't build content into your product, if you don't build user generated content into your product, you have nothing to train on or give these recommendations.”
Listen at 22:59
LLMs layered over ecommerce data frequently return hallucinated, irrelevant products.
“We put LLMs on top of e commerce data and it would hallucinate, hallucinate all the time and come back with products that were completely irrelevant.”
Listen at 25:28
Spacecraft fuel limits constrain payload capacity and travel range.
“all these spacecrafts are currently limited by the amount of fuel they can carry. So which compromises on the payload, how much payload they can carry, how much further they can go in space.”
Listen at 28:54
In-orbit refueling enables spacecraft to travel farther and carry more payload.
“So now they can travel further, carry more payload.”
Listen at 29:49
Spacecraft with insufficient fuel after a problem must currently abandon the mission and send another.
“Right now they just have to scrub that whole mission and send an A1.”
Listen at 31:07
Spacium has developed fuel tanks designed to prevent cryogenic propellant boil-off.
“we cracked that problem. We build zero boil off technology fuel tanks.”
Listen at 32:23
Cryogenic fuels provide more energy than less temperature-sensitive propellants.
“with crowding fields, you get more energy.”
Listen at 33:29
Electric, chemical, and nuclear propulsion systems can benefit from in-orbit refueling.
“electric, chemical, even nuclear propulsion will benefit from refueling”
Listen at 34:29
Spacium will need heavier launch vehicles when scaling fuel deliveries to 10–30 metric tons.
“when we have to scale up more like 10 to 30 metric tons of fuel, yes, then we will have to depend on heavier launch vehicles”
Listen at 35:41
Spacium’s current business plan does not depend on Starship or New Glenn succeeding.
“we are not depending on Starship or New Glenn for our business plan to work”
Listen at 35:51
Spacium’s in-house testing makes its hardware iteration cycle faster than typical space-industry processes.
“we brought all the testing in house. So our iteration cycle was really quick compared to how space industry usually works.”
Listen at 39:17
Spacium’s robotic arm is designed for positioning precision within 0.5 millimeters.
“End of the robotic arm will be we'll have a precision within 0.5 millimeters.”
Listen at 41:52
Spacium plans to transfer xenon in orbit within roughly the next year.
“We have a mission coming up really soon, within the next year or so. So we'll be transferring xenon in orbit”
Listen at 44:15
Spacecraft designs will change to exploit refueling within three to five years.
“we'll see that change coming in the, within the next three to five years.”
Listen at 47:26
Spacium reports nearly $100 million in contracts and more than $2 billion in letters of intent.
“we do have commercial contracts close to $100 million and we have two plus billion dollars in LOI.”
Listen at 48:44
Spacium expects profitability if its reported commercial contracts convert.
“once all those commercial contracts convert, we'll be able to be profitable.”
Listen at 48:52
SpaceX’s IPO increased public attention to the broader space industry.
“a lot of people started giving more attention to the space industry after SpaceX IPO.”
Listen at 50:14
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.
What people are saying
Why product search still needs more than an LLM
Wider topic conversation
- The episode’s topic centers on Onton’s neurosymbolic approach to making product search more reliable and personalized than chatbots.
Featuring
Mentioned
Listen to the full episode and explore every guest, topic, and moment on PodLume.