Our Technology
The approach is important.
{Apt・a}
Adjective
apt, suitable, fit (for), good (for), appropriate, adept
Agentic flow is the next
paradigm for conversational AI
Large Language Models e.g. Llama, Mistral, ChatGPT excel in their language capabilities but are held back due their inability to handle diverse data types and to conduct multi-step reasoning.
Unstructured Text Data
Majority of conversational copilots (LLM-based)
can only handle this
Large Structured Databases
Numerical
Charts & Diagrams
Multimodal
What's the weather today?
Correct Process: Identify the user's location ⟶ cross-reference the live weather in that location
​
LLM Process: Predicts the most likely next-word by modelling human language, without factoring in user location or live weather
In agentic flow, LLMs determine intent and formulate conversational outputs. Agents do the analytics and reasoning.
01
Problem
LLMs are designed to predict the next word as they are probabilistic models that are not optimized to perform logic or rule based tasks. Generalist LLMs lack the functionality to capture deep insights from large scale data.
02
Solution
We solve this by utilizing LLMs as the user facing mouth piece that then channel prompts to an array of bespoke expert models that are optimized for specific tasks.
03
Our Edge
Our technology builds on PhD research from the University of Cambridge. We develop a structure of highly functional collaborative AI experts through sophisticated prompt routing, task decomposition, and advanced methods for efficient model interface. Apta leverages niche expert systems to develop highly insightful features tailored to specific domains.
Generalist agentic flow has limitations for specialised applications.
Large Language Models e.g. Llama, Mistral, ChatGPT excel in their language capabilities but are held back due their inability to handle diverse data types and to conduct multi-step reasoning.
What are the attributes of US stocks that have performed the best in August 2024?
