A cacophonous but still considerate startup from Fresh York Augmented Intelligence Inc (AUI)which seeks to move beyond the popular “transformer” architecture used by most state-of-the-art LLM companies such as ChatGPT and Gemini, has raised $20 million in a SAFE bridge round at a valuation of $750 million, bringing total funding to nearly $60 millionVentureBeat can exclusively reveal.
The round, which will close in less than a week, comes amid increased interest in deterministic conversational AI and comes ahead of a larger raise that is currently in an advanced stage.
AUI is based on a combination of transformer technology and a newer technology called “neuro-symbolic artificial intelligence”, described in more detail below.
“We realize that you can combine the brilliance of the LLM in terms of language capabilities with the guarantees of symbolic artificial intelligence,” he said Good morning, Co-founder and CEO of AUI in a recent interview with VentureBeat. Elhelo founded the company in 2017 with her co-founder and chief product officer Ori Cohen.
The modern financing includes participation from eGateway Ventures, Fresh Era Capital Partners, existing shareholders and other strategic investors. This follows a $10 million raise in September 2024 at a $350 million valuation cap, coinciding with the company announced the launch of a market partnership with Google in October 2024. Early investors include Vertex Pharmaceuticals founder Joshua Boger, UKG CEO Aron Ain and former IBM CEO Jim Whitehurst.
According to the company, the bridge round is an announcement of a much larger escalate at an advanced stage.
AUI is the company behind Apollo-1, a modern core model built for task-oriented dialogue that it describes as the “economic half” of conversational AI – distinct from the open dialogue powered by LLMs such as ChatGPT and Gemini.
The company argues that existing LLMs lack the determinism, policy enforcement and operational certainty required by businesses, especially in regulated sectors.
Chris Varelas, co-founder of Redwood Capital and advisor to AUI, said in a press release provided to VentureBeat: “I’ve seen some of today’s top AI leaders walk away in a tizzy after interacting with Apollo-1.”
Characteristic neurosymbolic architecture
Apollo-1’s core innovation is its neurosymbolic architecture, which separates linguistic fluency from task-based reasoning. Instead of using the most popular technology that currently underpins most LLM and conversational AI systems – the infamous transformer architecture described in Google’s groundbreaking 2017 article “Attention Is All You Need” – the AUI system integrates two layers:
-
Neural modules powered by LLM handle perception: they encode user input and generate natural language responses.
-
The symbolic inference engine, developed over several years, interprets structured task elements such as intents, entities, and parameters. This symbolic state engine determines appropriate next actions using deterministic logic.
This hybrid architecture allows Apollo-1 to maintain state continuity, enforce organizational policies, and reliably trigger tool or API calls – capabilities that transformer-only agents lack.
Elhelo said this project was the result of years of data collection: “We created a consumer service and recorded millions of human-agent interactions between 60,000 active agents. From this we abstracted a symbolic language that defines the structure of task-based dialogs, separated from their domain-specific content.”
However, companies that have already built systems based on LLM transformers need not worry. AUI wants to make adopting modern technology just as effortless.
“Apollo-1 is deploying like any modern entry-level model,” Elhelo told VentureBeat in a text message last night. “It does not require dedicated or proprietary clusters to run. It runs in standard cloud and hybrid environments, leveraging both GPUs and CPUs, and is significantly cheaper to deploy than pioneering reasoning models. Apollo-1 can also be deployed across all major clouds in a separate environment for greater security.”
Generalization and domain flexibility
Apollo-1 is described as a basic task-oriented dialogue model, which means it is domain-agnostic and generalizable to industries such as healthcare, travel, insurance, and retail.
Unlike consultative AI platforms that require building logic tailored to each client’s needs, Apollo-1 enables enterprises to define behaviors and tools within a common symbolic language. This approach enables faster implementation and reduces long-term maintenance. According to the team, an enterprise can have a working agent up and running in less than a day.
Most importantly, procedural rules are encoded in the symbolic layer – not learned by example. This enables deterministic execution of sensitive or regulated tasks.
For example, the system can block the cancellation of a basic economy flight not by guessing, but by applying hard-coded logic to a symbolic representation of the booking class.
As Elhelo explained to VentureBeat, LLMs “are not a good mechanism when you’re looking for certainty. It’s better if you know what you’re sending [to an AI model] and always send it, and you always know what will come back [to the user] and how to deal with it.”
Availability and access for developers
Apollo-1 is already in active use in Fortune 500 companies in closed beta, with a broader, general release expected to be available by the end of 2025. previous report by Information, which broke the first news about the launch.
Enterprises can integrate with Apollo-1 by:
-
A developer playground where business users and technical teams collaborate to configure policies, rules, and behaviors; Or
-
Standard API using OpenAI compatible formats.
The model supports policy enforcement, rule-based customization, and guardrail control. Symbolic rules allow companies to dictate consistent behavior, while LLM modules handle open text interpretation and user interaction.
Enterprise Fit: When reliability outweighs smoothness
While LLMs are characterized by advanced dialogue and general-purpose creativity, they remain probabilistic – which is a barrier to enterprise implementation in finance, healthcare and customer service.
Apollo-1 fills this gap by offering a system in which rule adherence and deterministic task execution are primary design goals.
Elhelo puts it clearly: “If your use case is task-oriented dialogue, you must use us, even if you are ChatGPT.”