Back

Rahul Agarwal

Founder | Agentic AI...ย โ€ขย 22d

3 main types of memory used in AI systems. Iโ€™ve explained it in a simple way below. 1. ๐—˜๐—ฝ๐—ถ๐˜€๐—ผ๐—ฑ๐—ถ๐—ฐ ๐— ๐—ฒ๐—บ๐—ผ๐—ฟ๐˜† Episodic Memory is about ๐—ฟ๐—ฒ๐—บ๐—ฒ๐—บ๐—ฏ๐—ฒ๐—ฟ๐—ถ๐—ป๐—ด ๐—ฝ๐—ฎ๐˜€๐˜ ๐—ถ๐—ป๐˜๐—ฒ๐—ฟ๐—ฎ๐—ฐ๐˜๐—ถ๐—ผ๐—ป๐˜€ ๐—ฎ๐—ป๐—ฑ ๐—ฒ๐˜…๐—ฝ๐—ฒ๐—ฟ๐—ถ๐—ฒ๐—ป๐—ฐ๐—ฒ๐˜€. โ€ข ๐—ก๐—ผ๐˜๐—ถ๐—ฐ๐—ฒ ๐—ฒ๐˜ƒ๐—ฒ๐—ป๐˜: Detect when a new interaction happens (Eg: A user sends a message) โ€ข ๐—–๐—ผ๐—น๐—น๐—ฒ๐—ฐ๐˜ ๐—ฐ๐—ผ๐—ป๐˜๐—ฒ๐˜…๐˜: Capture details like time, user, and situation โ€ข ๐—ฅ๐—ฒ๐—ฐ๐—ผ๐—ฟ๐—ฑ ๐—ฎ๐—ฐ๐˜๐—ถ๐—ผ๐—ป: Save what the AI did in response โ€ข ๐—ฆ๐—ฎ๐˜ƒ๐—ฒ ๐—ผ๐˜‚๐˜๐—ฐ๐—ผ๐—บ๐—ฒ: Store the result of the interaction โ€ข ๐—–๐—ฟ๐—ฒ๐—ฎ๐˜๐—ฒ ๐—ฟ๐—ฒ๐—ฐ๐—ผ๐—ฟ๐—ฑ: Turn the event into a memory entry โ€ข ๐—ฅ๐—ฒ๐˜๐—ฟ๐—ถ๐—ฒ๐˜ƒ๐—ฒ ๐—ฟ๐—ฒ๐—ฐ๐—ผ๐—ฟ๐—ฑ๐˜€: Pull past memories when needed โ€ข ๐—–๐—ผ๐—บ๐—ฝ๐—ฎ๐—ฟ๐—ฒ ๐—ต๐—ถ๐˜€๐˜๐—ผ๐—ฟ๐˜†: Match new requests with past events โ€ข ๐—ข๐—ฟ๐—ด๐—ฎ๐—ป๐—ถ๐˜‡๐—ฒ ๐—ฑ๐—ฎ๐˜๐—ฎ: Structure memories for easy access โ€ข ๐—ฆ๐˜๐˜‚๐—ฑ๐˜† ๐—ฝ๐—ฎ๐˜๐˜๐—ฒ๐—ฟ๐—ป๐˜€: Detect repeated behaviors โ€ข ๐—š๐˜‚๐—ถ๐—ฑ๐—ฒ ๐—ฑ๐—ฒ๐—ฐ๐—ถ๐˜€๐—ถ๐—ผ๐—ป: Use past experience to improve responses โ€ข ๐—Ÿ๐—ผ๐—ด ๐—ป๐—ฒ๐˜„ ๐—ฟ๐—ฒ๐—ฐ๐—ผ๐—ฟ๐—ฑ: Store the latest interaction โ€ข ๐—˜๐˜…๐—ฝ๐—ฎ๐—ป๐—ฑ ๐—บ๐—ฒ๐—บ๐—ผ๐—ฟ๐˜†: Continuously grow memory over time ____________________ 2. ๐—ฆ๐—ฒ๐—บ๐—ฎ๐—ป๐˜๐—ถ๐—ฐ ๐— ๐—ฒ๐—บ๐—ผ๐—ฟ๐˜† Semantic Memory is about ๐—ธ๐—ป๐—ผ๐˜„๐—น๐—ฒ๐—ฑ๐—ด๐—ฒ, ๐—ณ๐—ฎ๐—ฐ๐˜๐˜€, ๐—ฎ๐—ป๐—ฑ ๐˜‚๐—ป๐—ฑ๐—ฒ๐—ฟ๐˜€๐˜๐—ฎ๐—ป๐—ฑ๐—ถ๐—ป๐—ด. โ€ข ๐—š๐—ฒ๐˜ ๐—พ๐˜‚๐—ฒ๐—ฟ๐˜†: Receive the userโ€™s question โ€ข ๐—œ๐—ฑ๐—ฒ๐—ป๐˜๐—ถ๐—ณ๐˜† ๐—ถ๐—ป๐˜๐—ฒ๐—ป๐˜: Understand what the user really means โ€ข ๐—ฆ๐—ฒ๐—ฎ๐—ฟ๐—ฐ๐—ต ๐—ธ๐—ป๐—ผ๐˜„๐—น๐—ฒ๐—ฑ๐—ด๐—ฒ: Look for relevant information โ€ข ๐—Ÿ๐—ถ๐—ป๐—ธ ๐—ฐ๐—ผ๐—ป๐—ฐ๐—ฒ๐—ฝ๐˜๐˜€: Connect related ideas together โ€ข ๐—ฃ๐˜‚๐—น๐—น ๐—ณ๐—ฎ๐—ฐ๐˜๐˜€: Gather accurate information โ€ข ๐—–๐—ผ๐—น๐—น๐—ฒ๐—ฐ๐˜ ๐—ฟ๐—ฒ๐—น๐—ฎ๐˜๐—ถ๐—ผ๐—ป๐˜€: Add supporting context โ€ข ๐—™๐—ผ๐—ฟ๐—บ๐—ฎ๐˜ ๐—ฟ๐—ฒ๐—ฝ๐—น๐˜†: Structure the response clearly โ€ข ๐—ฅ๐—ฒ๐—ณ๐—ฒ๐—ฟ๐—ฒ๐—ป๐—ฐ๐—ฒ ๐˜€๐—ผ๐˜‚๐—ฟ๐—ฐ๐—ฒ๐˜€: Use supporting knowledge if needed โ€ข ๐—ฆ๐—ฒ๐—ป๐—ฑ ๐—ผ๐˜‚๐˜๐—ฝ๐˜‚๐˜: Deliver the final answer โ€ข ๐— ๐—ฎ๐—ฟ๐—ธ ๐—ณ๐—ถ๐˜…๐—ฒ๐˜€: Learn from corrections โ€ข ๐—จ๐—ฝ๐—ฑ๐—ฎ๐˜๐—ฒ ๐—ธ๐—ป๐—ผ๐˜„๐—น๐—ฒ๐—ฑ๐—ด๐—ฒ: Improve knowledge over time ____________________ 3. ๐—ฃ๐—ฟ๐—ผ๐—ฐ๐—ฒ๐—ฑ๐˜‚๐—ฟ๐—ฎ๐—น ๐— ๐—ฒ๐—บ๐—ผ๐—ฟ๐˜† Procedural Memory is about ๐—ต๐—ผ๐˜„ ๐—”๐—œ ๐—ฝ๐—ฒ๐—ฟ๐—ณ๐—ผ๐—ฟ๐—บ๐˜€ ๐˜๐—ฎ๐˜€๐—ธ๐˜€ ๐—ฎ๐—ป๐—ฑ ๐˜„๐—ผ๐—ฟ๐—ธ๐—ณ๐—น๐—ผ๐˜„๐˜€. โ€ข ๐—š๐—ฒ๐˜ ๐˜๐—ฎ๐˜€๐—ธ: Receive a task to complete (Eg: Write an email) โ€ข ๐—œ๐—ฑ๐—ฒ๐—ป๐˜๐—ถ๐—ณ๐˜† ๐˜๐˜†๐—ฝ๐—ฒ: Understand the type of task โ€ข ๐— ๐—ฎ๐˜๐—ฐ๐—ต ๐˜๐—ผ ๐—ณ๐—น๐—ผ๐˜„: Select the correct workflow โ€ข ๐—Ÿ๐—ผ๐—ฎ๐—ฑ ๐—ฝ๐—ฟ๐—ผ๐—ฐ๐—ฒ๐˜€๐˜€: Prepare step-by-step instructions โ€ข ๐—”๐—ฑ๐—ฑ ๐—ถ๐—ป๐—ฝ๐˜‚๐˜๐˜€: Use user data and parameters โ€ข ๐—˜๐˜…๐—ฒ๐—ฐ๐˜‚๐˜๐—ฒ ๐˜€๐˜๐—ฒ๐—ฝ๐˜€: Perform actions one by one โ€ข ๐—ฆ๐—ฎ๐˜ƒ๐—ฒ ๐—ฝ๐—ฟ๐—ผ๐—ด๐—ฟ๐—ฒ๐˜€๐˜€: Track progress during execution โ€ข ๐—Ÿ๐—ผ๐—ด ๐—ผ๐˜‚๐˜๐—ฐ๐—ผ๐—บ๐—ฒ: Record results of the task โ€ข ๐—ฆ๐—ฒ๐—ป๐—ฑ ๐—ฟ๐—ฒ๐˜€๐˜‚๐—น๐˜: Deliver final output โ€ข ๐—›๐—ฎ๐—ป๐—ฑ๐—น๐—ฒ ๐—ฐ๐—ผ๐—ป๐—ฑ๐—ถ๐˜๐—ถ๐—ผ๐—ป๐˜€: Manage errors or special cases โ€ข ๐—–๐—ผ๐—ป๐˜๐—ถ๐—ป๐˜‚๐—ฒ ๐—ฝ๐—ฟ๐—ผ๐—ฐ๐—ฒ๐˜€๐˜€: Improve workflow for next time โœ… ๐—œ๐—ป ๐—ฆ๐—ต๐—ผ๐—ฟ๐˜: โ€ข Episodic โ†’ remembers events & experiences โ€ข Semantic โ†’ understands knowledge & meaning โ€ข Procedural โ†’ knows how to perform tasks Very important for people to understand who are building AI systems and processes. โœ… Repost for others so they can understand different memory types in AI.

Reply
1
4

More like this

Recommendations from Medial

Image Description

Rahul Agarwal

Founder | Agentic AI...ย โ€ขย 6m

How Multi-Agent AI systems actually work? Explained in a very simple way. Read below: -> ๐—ง๐—ต๐—ฒ ๐— ๐—ฎ๐—ถ๐—ป ๐—”๐—œ ๐—”๐—ด๐—ฒ๐—ป๐˜ The main ๐—”๐—œ ๐—”๐—ด๐—ฒ๐—ป๐˜ is the ๐—ผ๐—ฟ๐—ฐ๐—ต๐—ฒ๐˜€๐˜๐—ฟ๐—ฎ๐˜๐—ผ๐—ฟ. It has several capabilities: โ€ข ๐——๐—ฎ๐˜๐—ฎ๐—ฏ๐—ฎ๐˜€๐—ฒ โ€“ Stores knowledge o

See More
Reply
6
19
1

Rahul Agarwal

Founder | Agentic AI...ย โ€ขย 1m

Very few people understand, how AI memory works. I've explained in simple below. ๐—ฆ๐—ต๐—ผ๐—ฟ๐˜-๐—ง๐—ฒ๐—ฟ๐—บ ๐— ๐—ฒ๐—บ๐—ผ๐—ฟ๐˜†(๐—ฆ๐—ง๐— ) 1. ๐—”๐—ฐ๐—ฐ๐—ฒ๐—ฝ๐˜ ๐—ถ๐—ป๐—ฝ๐˜‚๐˜ โ€“The system receives your message (question/prompt). 2. ๐—•๐—ฟ๐—ฒ๐—ฎ๐—ธ ๐—ถ๐—ป๐˜๐—ผ ๐˜๐—ผ๐—ธ๐—ฒ๐—ป๐˜€โ€“ Your mes

See More
Reply
1

Rahul Agarwal

Founder | Agentic AI...ย โ€ขย 9d

Everyone should know how Agentic AI really works. Iโ€™ve explained it in a simple way below. 1. ๐—จ๐˜€๐—ฒ๐—ฟ ๐—œ๐—ป๐˜๐—ฒ๐—ฟ๐—ณ๐—ฎ๐—ฐ๐—ฒ (๐—™๐—ฟ๐—ผ๐—ป๐˜๐—ฒ๐—ป๐—ฑ) Everything begins with the ๐—จ๐˜€๐—ฒ๐—ฟ ๐—œ๐—ป๐˜๐—ฒ๐—ฟ๐—ณ๐—ฎ๐—ฐ๐—ฒ. โ€ข Users type a message or give a task โ€ข This happen

See More
Reply
1

Rahul Agarwal

Founder | Agentic AI...ย โ€ขย 3m

Hands down the simplest explanation of AI agents using LLMs, memory, and tools. A user sends an input โ†’ the system (agent) builds a prompt and may call tools and memory-search (RAG) โ†’ agent decides and builds an answer โ†’ the answer is returned to th

See More
Reply
2
7

Rahul Agarwal

Founder | Agentic AI...ย โ€ขย 1m

Most non-tech people learning AI donโ€™t get this. I've explained it in a simple way below. 1. ๐—จ๐˜€๐—ฒ๐—ฟ Everything starts with the ๐—จ๐˜€๐—ฒ๐—ฟ. โ€ข The user wants something done โ€ข Example: โ€œ๐˜๐˜ช๐˜ฏ๐˜ฅ ๐˜ต๐˜ฉ๐˜ฆ ๐˜ฃ๐˜ฆ๐˜ด๐˜ต ๐˜ญ๐˜ข๐˜ฑ๐˜ต๐˜ฐ๐˜ฑ ๐˜ถ๐˜ฏ๐˜ฅ๐˜ฆ๐˜ณ $1000โ€ or โ€œ๐˜ž๐˜ณ๏ฟฝ

See More
Reply
1
4

Rahul Agarwal

Founder | Agentic AI...ย โ€ขย 24d

If you think an AI agent is just a smarter chatbot, youโ€™re already behind. An agent is a coordinated system that perceives, decides, and executes. Thatโ€™s a different engineering problem. In production, the agent connects your data and external envi

See More
Reply
1

Rahul Agarwal

Founder | Agentic AI...ย โ€ขย 2m

Most people building modern AI systems miss these steps. I've explained each step in a simple way below. 1. ๐— ๐˜‚๐—น๐˜๐—ถ-๐—”๐—ด๐—ฒ๐—ป๐˜ ๐—œ๐—ป๐˜๐—ฒ๐—ฟ๐—ผ๐—ฝ๐—ฒ๐—ฟ๐—ฎ๐—ฏ๐—ถ๐—น๐—ถ๐˜๐˜† How multiple AI agents work together as a system. Step-by-step: โ€ข ๐—จ๐˜€๐—ฒ๐—ฟ ๐—ฅ๐—ฒ๐—พ๐˜‚๏ฟฝ

See More
2 Replies
1
8

Rahul Agarwal

Founder | Agentic AI...ย โ€ขย 2m

4 powerful loops that power Agentic AI. Hereโ€™s the easiest explanation of how each one works. ๐—”๐—š๐—˜๐—ก๐—ง๐—œ๐—– ๐—Ÿ๐—ข๐—ข๐—ฃ๐—ฆ Agentic Loops explain how AI agents think, act, learn, coordinate, and improve over time using structured cycles. 1. ๐—–๐—ผ๐—น๐—น๐—ฎ

See More
Reply
2
7
Image Description

Rahul Agarwal

Founder | Agentic AI...ย โ€ขย 3m

LangGraph vs Autogen vs CrewAI, which to choose? I've given a breakdown of which is best for you. ๐—Ÿ๐—ฎ๐—ป๐—ด๐—š๐—ฟ๐—ฎ๐—ฝ๐—ต (๐—–๐—ผ๐—บ๐—ฝ๐—น๐—ฒ๐˜… ๐—ช๐—ผ๐—ฟ๐—ธ๐—ณ๐—น๐—ผ๐˜„๐˜€) LangGraph is ideal for ๐—ฏ๐˜‚๐—ถ๐—น๐—ฑ๐—ถ๐—ป๐—ด ๐—ฐ๐—ผ๐—บ๐—ฝ๐—น๐—ฒ๐˜…, ๐—บ๐˜‚๐—น๐˜๐—ถ-๐˜€๐˜๐—ฒ๐—ฝ ๐—”๐—œ ๐˜„๐—ผ๐—ฟ๐—ธ๐—ณ๐—น๐—ผ

See More
Reply
12
1

Rahul Agarwal

Founder | Agentic AI...ย โ€ขย 3m

Everyone wants to build AI agents these days. But very few actually understand what sits beneath the surface. Hereโ€™s the part most people ignore: AI agents are mostly software engineering - about 95%. The โ€œAIโ€ part is just the remaining 5%. All the

See More
Reply
3

Download the medial app to read full posts, comements and news.