TILs - Fueling Curiosity, One Insight at a Time

At Codemancers, we believe every day is an opportunity to grow. This section is where our team shares bite-sized discoveries, technical breakthroughs and fascinating nuggets of wisdom we've stumbled upon in our work.

Published: Mar 7, 2025
Author: Nived Hari
System Analyst

You can manually send messages to a Kafka topic using Karafka's producer. This is useful for debugging, testing, or custom event handling.
Example:

Ruby

payload = {
  id: 123,
  name: "Sample Item",
  status: "processed",
  timestamp: 
Time.now.to_i
}

Karafka.producer.produce_sync(
  topic: "your_topic_name",
  payload: payload.to_json
)

Key Points:
• produce_sync ensures the message is sent before proceeding.
• topic specifies the Kafka topic where the message will be published.
• payload should be serialized into JSON or another supported format.
#karafka

Published: Mar 6, 2025
Author: Nitturu Baba
System Analyst

Searching in vector databases

1️⃣ Convert Text to Embeddings
• Text is transformed into numerical vectors using AI models like OpenAI, BERT, or Sentence Transformers.
2️⃣ Index & Organise Embeddings
• Instead of scanning all vectors, the database groups similar embeddings into clusters (buckets) to speed up search.
• Common indexing methods:
◦ HNSW (Hierarchical Navigable Small World) – builds a graph where similar embeddings are connected, reducing search time.
◦ IVFFLAT (Inverted File Index) – divides embeddings into clusters (buckets) and searches only the most relevant ones.
3️⃣ Search Using Similarity Metrics
• The query is converted into an embedding and compared to stored vectors using:
◦ Cosine Similarity: Cosine Similarity measures the angle between vectors while ignoring their magnitude, where a higher value means greater similarity (1 = identical, 0 = unrelated, -1 = opposite). It is commonly used for text similarity, such as document searches.
◦ Euclidean Distance: Euclidean Distance calculates the straight-line distance between points, where a lower value means greater similarity (0 = identical). This method is ideal for spatial data, like image or geographical searches.
• The database searches only the closest clusters, making it faster.
4️⃣ Return the Closest Matches
• The best matches (top K documents) are ranked and returned based on similarity scores.
📌 Convert text → embeddings, group them into clusters, search only relevant ones, return the top K ranked results.

#vectordatabase

Published: Mar 5, 2025
Author: Nitturu Baba
System Analyst

RAG has three key steps:
1️⃣ Retrieval – Fetch relevant context from a vector database.
2️⃣ Augmentation – Inject the retrieved context into the prompt.
3️⃣ Generation – Use an LLM (GPT, Llama, etc.) to produce a fact-based response.

🔹 Step 1: Retrieval – Finding Relevant Information
Before answering a question, the system searches for relevant documents in a vector database.
💬 Example Question: "What is the capital of France?"
🔍 Retrieval Process:
• The system searches for relevant text in a vector database.
• It finds a stored Wikipedia snippet:
Paris is the capital of France, known for the Eiffel Tower.

📌 Retrieved Context:
Paris is the capital of France, known for the Eiffel Tower.

🔹 Step 2: Augmentation – Enriching the Prompt with Context
After retrieving relevant information, the system adds it to the prompt.

📌 Final Augmented Prompt:
User Question: "What is the capital of France?"
Retrieved Context: "Paris is the capital of France, known for the Eiffel Tower."
Final Prompt: "Using the provided context, answer: What is the capital of France?"

👉 Why is this useful?
✅ Retrieval ensures AI has up-to-date context instead of relying only on pre-trained data.
✅ Augmentation refines the LLM’s input, making answers more precise.
✅ Reduces hallucinations, ensuring the AI doesn’t generate incorrect facts.

🔹 Step 3: Generation – Producing the Final Answer
Once the AI has retrieved and augmented the prompt, it generates a final response.
💡 Example Output:
"The capital of France is Paris, known for the Eiffel Tower and rich history."

#AI #RAG

Published: Mar 4, 2025
Author: Adithya Hebbar
System Analyst

Updating Session in NextAuth

In NextAuth, you can update the session data using the update function from useSession(). Here's how you can modify user details dynamically:

JavaScript

const { data: session, update } = useSession();

await update({
  user: {
    ...session?.user,
    name: "Updated Name",
    role: "editor", 
  },
});

Assuming a strategy: "jwt" is used, the update() method will trigger a jwt callback with the trigger: "update" option. You can use this to update the session object on the server.

Code

export default NextAuth({
  callbacks: {
    // Using the `...rest` parameter to be able to narrow down the type based on `trigger`
    jwt({ token, trigger, session }) {
      if (trigger === "update" && session?.name) {
        // Note, that `session` can be any arbitrary object, remember to validate it!
        token.name = session.name
        token.role = session.role
      }
      return token
    }
  }
})

This updates the session without requiring a full reload, ensuring the UI reflects the changes immediately. 🚀

#next-auth #nextjs

Published: Feb 25, 2025
Author: Puneeth kumar
System Analyst

Traits in FactoryBot helps to define reusable variations of a factory without creating multiple factories. They are useful when we need optional attributes or specific states in test data.
Let's say we have a User model with different roles (admin, regular, guest). Instead of writing separate factories, we can use traits like below:

Ruby

# spec/factories/users.rb
FactoryBot.define do
  factory :user do
    first_name { Faker::Name.first_name }
    email { Faker::Internet.unique.email }
    password { "password123" }

    trait :admin do
      role { "admin" }
    end

    trait :guest do
      role { "guest" }
    end

    trait :confirmed do
      confirmed_at { Time.current }
    end
  end
end

And use it like below

Code

let(:admin_user) { create(:user, :admin) }
let(:guest_user) { create(:user, :guest) }
let(:confirmed_user) { create(:user, :confirmed) }

#CU6U0R822 #factory_bot

Published: Feb 25, 2025
Author: Nived Hari
System Analyst

In Ruby, public_send allows calling a method dynamically when its name is stored in a variable.

• Why Use public_send?
Instead of calling methods explicitly, we can determine the method name at runtime and call it dynamically.

Example: Handling Different Attribute Names

Ruby

quantity_field = item.respond_to?(:ordered_quantity) ? :ordered_quantity : :quantity
new_quantity = item.public_send(quantity_field).to_i + item_case[:quantity].to_i

• If item has ordered_quantity, it calls item.ordered_quantity
• Otherwise, it calls item.quantity
• This avoids unnecessary if-else statements
#ruby

Published: Feb 25, 2025
Author: Nived Hari
System Analyst

In Ruby, there are two ways to define hash keys:

1. Using the Colon Syntax (:) – Creates a Literal Symbol Key

Ruby

item.update!(
  ordered_quantity: new_quantity,
)

Key Behavior: The key is treated as a fixed symbol (e.g., :ordered_quantity).

2. Using the Hash Rocket (=>) – Evaluates the Left-Hand Side as a Key

Code

item.update!(
  quantity_field => new_quantity,
)

Key Behavior: The left-hand side is evaluated dynamically, making it useful for variable-based keys.

Example Use Case: Dynamic Keys

Code

quantity_field = item.respond_to?(:ordered_quantity) ? :ordered_quantity : :quantity
new_quantity = item.public_send(quantity_field).to_i + item_case[:quantity].to_i

item.update!(
  quantity_field => new_quantity,  # Evaluates to :ordered_quantity or :quantity
)

Here, quantity_field is determined dynamically based on the model, so => must be used instead of :.

When to Use =>?
• When working with multiple models that have different column names
• When dynamically generating hash keys at runtime
• When building flexible APIs that handle varying attribute names
Takeaway:
• Use : when the key is static and always the same.
• Use => when the key is stored in a variable or needs to be evaluated dynamically.
#ruby

Published: Feb 20, 2025
Author: Nived Hari
System Analyst

In Ruby, attr_reader automatically creates a getter method for instance variables, making code cleaner and more concise. Instead of writing:

Ruby

def some_number
  @some_number
end

You can simply use:

Code

attr_reader :some_number

This makes attributes read-only while keeping the class lightweight

#ruby #CU6U0R822

Published: Feb 20, 2025
Author: Nitturu Baba
System Analyst

Real-time AI response streaming improves user experience by reducing wait times and making interactions feel more dynamic. Instead of waiting for the entire response to be generated before displaying it, streaming allows data to be processed and presented incrementally.

Example of AI response streaming using Nest Js backend and Next JS front end.

Setting Up the NestJS Backend for Streaming AI Responses

Controller

TypeScript

import { Controller, Post, Body, Res } from '@nestjs/common';
import { openai } from '@ai-sdk/openai';
import { streamText } from 'ai';
import { Response } from 'express';

@Controller('orchestrator')
export class OrchestratorController {

 @Post('chat')
 async chat(@Body() payload: any, @Res() res: Response) {
  const { messages } = payload;

  const result = streamText({
   model: openai('gpt-4o'),
   messages,
  });
   
  result.pipeDataStreamToResponse(res); // Streams the AI response directly to the client
 }
}

How It Works
• The @Post('chat') endpoint listens for chat requests.
• The streamText function sends user messages to OpenAI and receives a streamed response.
• pipeDataStreamToResponse(res) directly streams the AI-generated content to the client as it arrives.
Building the Next.js Frontend for AI Response Streaming

chat/page.tsx

JavaScript

'use client';

import { useChat } from '@ai-sdk/react';

export default function Home() {
 const { messages, input, handleInputChange, handleSubmit } = useChat({
  api: 'https://localhost:3000/api/orchestrator/chat',  // make the post request to the NestJS backend
 });

 return (
  <div>
   {messages.map((message) => (
    <div key={message.id}>
     {message.role === 'user' ? 'User: ' : 'AI: '}
     {message.content}
    </div>
   ))}

   <form onSubmit={handleSubmit}>
    <input
     name="prompt"
     value={input}
     onChange={handleInputChange}
     className="text-black"
    />
    <button type="submit">Submit</button>
   </form>
  </div>
 );
}

How It Works
• The useChat hook from AI-SDK manages state and streaming logic automatically.
• It sends user messages to the backend and updates the UI in real time as responses arrive.
• The messages array dynamically updates, displaying each chunk of AI-generated text as it's received.
#C08DPTN3JAW #streaming #next js #nest js

Published: Feb 20, 2025
Author: Puneeth kumar
System Analyst

Collection caching is a way to speed up rendering multiple items on a page by storing their HTML in cache.
How it works:
1. When we use <%= render partial: 'products/product', collection: @products, cached: true %>, Rails checks if each product's HTML is already stored in the cache.
2. If a product’s HTML is found in the cache, Rails loads it quickly instead of rendering it again.
3. If a product’s HTML is not in the cache, Rails will render it, store it in the cache, and use it next time.
4. The big advantage: Rails fetches all cached products at once (instead of one by one), making it much faster.
#CU6U0R822 #caching #collection_caching

Showing page 9 of 83

Ready to Build Something Amazing?

Codemancers can bring your vision to life and help you achieve your goals

Let's Talk

Address
2nd Floor, Zee Plaza,
No. 1678, 27th Main Rd,
Sector 2, HSR Layout,
Bengaluru, Karnataka 560102
Contact
hello@codemancers.com
+91-9731601276