From: aidotengineer
LinkedIn’s Generative AI (GenAI) platform evolved significantly to support complex AI applications, with key investments made in areas like tool calling (or skills invocation) and memory management [00:05:05], [00:05:08]. The platform classifies these components into four layers: orchestration, prompt engineering, tools and skills invocation, and content and memory management [00:07:44], [00:07:47], [00:07:50], [00:07:53].
Skills Invocation (Tools and Skills)
In the era of AI agents, skills or APIs are a crucial aspect, as agents are expected to perform specific actions [00:05:36], [00:05:39], [00:05:42]. LinkedIn made a significant investment in a “skill registry” to facilitate this [00:05:48].
Skill Registry
The skill registry provides a set of tools that allow developers to publish their APIs into a centralized repository [00:05:51], [00:05:54], [00:05:56]. This registry addresses key challenges:
- Skill Discovery [00:06:00]
- Skill Invocation [00:06:03]
By centralizing skills, it becomes very easy for applications to call APIs and perform tasks [00:06:05], [00:06:07], [00:06:10].
The ability to uplift APIs into skills that can be easily called by agents is considered a critical new component for the agent era, requiring surrounding tooling and infrastructure support [00:15:42], [00:15:46], [00:15:49], [00:15:53], [00:15:56], [00:15:58]].
Memory Management
Memory management is a crucial component for injecting rich data into the agent experience [00:15:26], [00:15:30], [00:15:32], [00:15:34], [00:15:38].
Conversational Memory
Initial platform capabilities included “conversational memory” infrastructure, which helps track LLM interactions and retrieval content. This content is then injected into the final product to enable conversational bots [00:04:08], [00:04:11], [00:04:14], [00:04:17], [00:04:20], [00:04:23], [00:04:26].
Experiential Memory
Beyond conversational memory, the platform expanded its capabilities to include “experiential memory” [00:06:16], [00:06:18], [00:06:21]. This is a memory storage system designed to extract, analyze, and infer factual knowledge from interactions between the agent and the user [00:06:24], [00:06:28], [00:06:31], [00:06:32].
Memory is organized into different layers to help agents be aware of surrounding content [00:06:35], [00:06:39], [00:06:42], [00:06:45], [00:06:47]. These layers include:
- Working memory [00:06:39]
- Long-term memory [00:06:42]
- Collective memories [00:06:42]
LinkedIn has found success leveraging its existing messaging infrastructure to serve as a cost-efficient and scalable memory layer [00:16:31], [00:16:33], [00:16:36], [00:16:39], [00:16:42].