Samik
@prosamik.bsky.social
Let's Build and Ship SaaS Together
No Course. No Pitch
Just Honest execution
No Course. No Pitch
Just Honest execution
Repeated action based on your habits compounds the results you achieve.
I started by building my portfolio- prosamik.com
Now, I've scaled that action to build my own AI SaaS- mapyourideas.com
That's why
Repeated Action > Multiple Actions
I started by building my portfolio- prosamik.com
Now, I've scaled that action to build my own AI SaaS- mapyourideas.com
That's why
Repeated Action > Multiple Actions
May 8, 2025 at 1:33 PM
Repeated action based on your habits compounds the results you achieve.
I started by building my portfolio- prosamik.com
Now, I've scaled that action to build my own AI SaaS- mapyourideas.com
That's why
Repeated Action > Multiple Actions
I started by building my portfolio- prosamik.com
Now, I've scaled that action to build my own AI SaaS- mapyourideas.com
That's why
Repeated Action > Multiple Actions
Think your Agent is safe without guardrails?
That’s like leaving your front door wide open.
7 types of guardrails for your Agent 👇
That’s like leaving your front door wide open.
7 types of guardrails for your Agent 👇
May 7, 2025 at 1:26 PM
Think your Agent is safe without guardrails?
That’s like leaving your front door wide open.
7 types of guardrails for your Agent 👇
That’s like leaving your front door wide open.
7 types of guardrails for your Agent 👇
This is a simple summarisation engine which summarises a YouTube video.
How does this work?
1/ Form trigger, which takes a YouTube video URL
2/ It calls the Apify actor to fetch the video transcript.
3/ Then it calls GPT 4.1 for a concise transcript summary.
4/ And match with different columns.
How does this work?
1/ Form trigger, which takes a YouTube video URL
2/ It calls the Apify actor to fetch the video transcript.
3/ Then it calls GPT 4.1 for a concise transcript summary.
4/ And match with different columns.
May 2, 2025 at 12:15 PM
This is a simple summarisation engine which summarises a YouTube video.
How does this work?
1/ Form trigger, which takes a YouTube video URL
2/ It calls the Apify actor to fetch the video transcript.
3/ Then it calls GPT 4.1 for a concise transcript summary.
4/ And match with different columns.
How does this work?
1/ Form trigger, which takes a YouTube video URL
2/ It calls the Apify actor to fetch the video transcript.
3/ Then it calls GPT 4.1 for a concise transcript summary.
4/ And match with different columns.
The simplest way to understand an n8n workflow is to create one.
And I've done the same for you.
Getting started is what matters
And I've done the same for you.
Getting started is what matters
May 2, 2025 at 12:15 PM
The simplest way to understand an n8n workflow is to create one.
And I've done the same for you.
Getting started is what matters
And I've done the same for you.
Getting started is what matters
Stop Thinking
Start Executing.
Thinking is fragile.
Whenever I do this, I don't see any results.
Execution gives you the real results.
Start Executing.
Thinking is fragile.
Whenever I do this, I don't see any results.
Execution gives you the real results.
April 28, 2025 at 11:42 AM
Stop Thinking
Start Executing.
Thinking is fragile.
Whenever I do this, I don't see any results.
Execution gives you the real results.
Start Executing.
Thinking is fragile.
Whenever I do this, I don't see any results.
Execution gives you the real results.
If you’re on Day 0 of building a startup,
And you’ve 0 intention of spending money on emails for your first 100 users
Then, I’ve a tip for you 👇
And you’ve 0 intention of spending money on emails for your first 100 users
Then, I’ve a tip for you 👇
April 27, 2025 at 1:28 PM
If you’re on Day 0 of building a startup,
And you’ve 0 intention of spending money on emails for your first 100 users
Then, I’ve a tip for you 👇
And you’ve 0 intention of spending money on emails for your first 100 users
Then, I’ve a tip for you 👇
8/ Claude Computer is an example of an Agent:
When developing Agents. The autonomous nature of agents means higher costs, and the potential for compounding errors. They recommend extensive testing in sandboxed environments, along with the appropriate guardrails.
When developing Agents. The autonomous nature of agents means higher costs, and the potential for compounding errors. They recommend extensive testing in sandboxed environments, along with the appropriate guardrails.
April 26, 2025 at 3:27 PM
8/ Claude Computer is an example of an Agent:
When developing Agents. The autonomous nature of agents means higher costs, and the potential for compounding errors. They recommend extensive testing in sandboxed environments, along with the appropriate guardrails.
When developing Agents. The autonomous nature of agents means higher costs, and the potential for compounding errors. They recommend extensive testing in sandboxed environments, along with the appropriate guardrails.
7/ Agents:
Agents, on the other hand, are used for open-ended problems, where it’s difficult to predict the required number of steps to perform a specific task by hardcoding the steps.
Agents need autonomy over the environment, and you have to trust their decision-making to some extent.
Agents, on the other hand, are used for open-ended problems, where it’s difficult to predict the required number of steps to perform a specific task by hardcoding the steps.
Agents need autonomy over the environment, and you have to trust their decision-making to some extent.
April 26, 2025 at 3:27 PM
7/ Agents:
Agents, on the other hand, are used for open-ended problems, where it’s difficult to predict the required number of steps to perform a specific task by hardcoding the steps.
Agents need autonomy over the environment, and you have to trust their decision-making to some extent.
Agents, on the other hand, are used for open-ended problems, where it’s difficult to predict the required number of steps to perform a specific task by hardcoding the steps.
Agents need autonomy over the environment, and you have to trust their decision-making to some extent.
6/ Workflow: Evaluator-optimizer
We use this when we have some evaluation criteria for the result, and with refinement through iteration, it provides measurable value
You can put a human in the loop for evaluation, or let LLM decide feedback dynamically
We use this when we have some evaluation criteria for the result, and with refinement through iteration, it provides measurable value
You can put a human in the loop for evaluation, or let LLM decide feedback dynamically
April 26, 2025 at 3:27 PM
6/ Workflow: Evaluator-optimizer
We use this when we have some evaluation criteria for the result, and with refinement through iteration, it provides measurable value
You can put a human in the loop for evaluation, or let LLM decide feedback dynamically
We use this when we have some evaluation criteria for the result, and with refinement through iteration, it provides measurable value
You can put a human in the loop for evaluation, or let LLM decide feedback dynamically
5/ Workflow: Orchestrator-workers
Similar to Parallelization, but here the sub-tasks are decided by the LLM dynamically.
In the Final step, the results are aggregated into one.
Best example:
Coding Products that make complex changes to multiple files each time.
Similar to Parallelization, but here the sub-tasks are decided by the LLM dynamically.
In the Final step, the results are aggregated into one.
Best example:
Coding Products that make complex changes to multiple files each time.
April 26, 2025 at 3:27 PM
5/ Workflow: Orchestrator-workers
Similar to Parallelization, but here the sub-tasks are decided by the LLM dynamically.
In the Final step, the results are aggregated into one.
Best example:
Coding Products that make complex changes to multiple files each time.
Similar to Parallelization, but here the sub-tasks are decided by the LLM dynamically.
In the Final step, the results are aggregated into one.
Best example:
Coding Products that make complex changes to multiple files each time.
4/ Workflow: Parallelization
Done in two formats:
Section-wise: Breaking a complex task into subtasks and combining all results in one place
Voting: Running the same task multiple times and selecting the final output based on ranking
Done in two formats:
Section-wise: Breaking a complex task into subtasks and combining all results in one place
Voting: Running the same task multiple times and selecting the final output based on ranking
April 26, 2025 at 3:27 PM
4/ Workflow: Parallelization
Done in two formats:
Section-wise: Breaking a complex task into subtasks and combining all results in one place
Voting: Running the same task multiple times and selecting the final output based on ranking
Done in two formats:
Section-wise: Breaking a complex task into subtasks and combining all results in one place
Voting: Running the same task multiple times and selecting the final output based on ranking
3/ Workflow: Routing
Best Example:
Customer support where you route different queries for different services
Best Example:
Customer support where you route different queries for different services
April 26, 2025 at 3:27 PM
3/ Workflow: Routing
Best Example:
Customer support where you route different queries for different services
Best Example:
Customer support where you route different queries for different services
2/ Workflow: Prompt Chaining
Here, different LLMs are performing a specific task in a series, and Gate verifies the output of each LLM call
Best example:
Generating a Marketing Copy with your style and then converting it into different Languages
Here, different LLMs are performing a specific task in a series, and Gate verifies the output of each LLM call
Best example:
Generating a Marketing Copy with your style and then converting it into different Languages
April 26, 2025 at 3:27 PM
2/ Workflow: Prompt Chaining
Here, different LLMs are performing a specific task in a series, and Gate verifies the output of each LLM call
Best example:
Generating a Marketing Copy with your style and then converting it into different Languages
Here, different LLMs are performing a specific task in a series, and Gate verifies the output of each LLM call
Best example:
Generating a Marketing Copy with your style and then converting it into different Languages
1/ The Foundational Building Block
Augmented LLM:
The basic building block of agentic systems is an LLM enhanced with augmentations such as retrieval, tools, and memory
The best example of Augmented LLM is Model Context Protocol (MCP)
Augmented LLM:
The basic building block of agentic systems is an LLM enhanced with augmentations such as retrieval, tools, and memory
The best example of Augmented LLM is Model Context Protocol (MCP)
April 26, 2025 at 3:27 PM
1/ The Foundational Building Block
Augmented LLM:
The basic building block of agentic systems is an LLM enhanced with augmentations such as retrieval, tools, and memory
The best example of Augmented LLM is Model Context Protocol (MCP)
Augmented LLM:
The basic building block of agentic systems is an LLM enhanced with augmentations such as retrieval, tools, and memory
The best example of Augmented LLM is Model Context Protocol (MCP)
What you think is an Agent is actually a Workflow
People behind Claude say it is an Agentic System
🧵Simplified Version of Anthropic’s guide
Understand different Agentic Architectural Patterns here 👇
People behind Claude say it is an Agentic System
🧵Simplified Version of Anthropic’s guide
Understand different Agentic Architectural Patterns here 👇
April 26, 2025 at 3:27 PM
What you think is an Agent is actually a Workflow
People behind Claude say it is an Agentic System
🧵Simplified Version of Anthropic’s guide
Understand different Agentic Architectural Patterns here 👇
People behind Claude say it is an Agentic System
🧵Simplified Version of Anthropic’s guide
Understand different Agentic Architectural Patterns here 👇
Windsurf has turned the tables for Cursor by changing its pricing models
Say goodbye to complex pricing confusion.
🧵 Here are the 7 things you need to know:
Say goodbye to complex pricing confusion.
🧵 Here are the 7 things you need to know:
April 23, 2025 at 11:00 AM
Windsurf has turned the tables for Cursor by changing its pricing models
Say goodbye to complex pricing confusion.
🧵 Here are the 7 things you need to know:
Say goodbye to complex pricing confusion.
🧵 Here are the 7 things you need to know:
I've been using Cursor for 90 days
I’d have saved a lot of time if someone had told me these 7 things
I’d have saved a lot of time if someone had told me these 7 things
April 21, 2025 at 11:43 AM
I've been using Cursor for 90 days
I’d have saved a lot of time if someone had told me these 7 things
I’d have saved a lot of time if someone had told me these 7 things
My personal favourite is Claude for Agentic Purposes. What’s your favourite?
Check out this registry, which I find cool-
Check out this registry, which I find cool-
April 18, 2025 at 2:51 PM
My personal favourite is Claude for Agentic Purposes. What’s your favourite?
Check out this registry, which I find cool-
Check out this registry, which I find cool-
The best habit I made in 2025:
Coding each day, even if it's only one commit
Here’s what resulted from the Consistent building:
I built 5 tools for Solopreneurs looking to save time
Coding each day, even if it's only one commit
Here’s what resulted from the Consistent building:
I built 5 tools for Solopreneurs looking to save time
April 18, 2025 at 6:04 AM
The best habit I made in 2025:
Coding each day, even if it's only one commit
Here’s what resulted from the Consistent building:
I built 5 tools for Solopreneurs looking to save time
Coding each day, even if it's only one commit
Here’s what resulted from the Consistent building:
I built 5 tools for Solopreneurs looking to save time