# Saurabh's Blog — full content

> The playbook for delivering digital products. Processes, tools, and frameworks — simplified from real experience.

Site: https://saurabh.dev
Generated from 4 published posts. Raw markdown bodies are
included verbatim under their post's metadata header.

## Journeys
- Technical Concepts: The building blocks you need to understand — networking, architecture, databases, and the systems that power digital products. (https://saurabh.dev/journeys/technical-concepts/)
- Decision Frameworks: How to think through hard problems — frameworks for ambiguity, trade-offs, and making decisions that hold up. (https://saurabh.dev/journeys/decision-frameworks/)
- AI in Practice: Using AI to deliver products — decision frameworks, integration patterns, and practical applications from the field. (https://saurabh.dev/journeys/ai-in-practice/)
- Delivery Discipline: From concept to production — how to ship, scale, and maintain digital products that work. (https://saurabh.dev/journeys/delivery-discipline/)
- Exploration: Hands-on tutorials and practical guides — learn by doing with real-world tools and workflows. (https://saurabh.dev/journeys/exploration/)
- Reflections: Lessons from the field — insights about shipping, leading teams, and building things that work. (https://saurabh.dev/journeys/reflections/)

---

# Running Llama 3 Locally: A Privacy-First AI Guide
URL: https://saurabh.dev/blog/run-llama-locally/
Published: 2026-07-08
Journey: AI in Practice
Tags: tech, llama, ollama, privacy, llm

## Why Run AI Locally?

Cloud AI services are convenient, but they come with trade-offs:

- **Privacy** — Your data leaves your network
- **Cost** — API calls add up quickly
- **Latency** — Network round-trips add delay
- **Dependency** — What if the service goes down?

Running models locally solves all of these issues.

## Hardware Requirements

### Minimum Setup

- **CPU** — Modern multi-core (Intel 12th gen or AMD Ryzen 5000+)
- **RAM** — 16GB (for 7B models)
- **GPU** — Optional but recommended (NVIDIA with 8GB+ VRAM)

### Recommended Setup

- **CPU** — Intel i7/i9 or AMD Ryzen 7/9
- **RAM** — 32GB+
- **GPU** — NVIDIA RTX 3060 12GB or better

## Installing Ollama

Ollama is the easiest way to run LLMs locally:

```bash
# macOS / Linux
curl -fsSL https://ollama.com/install.sh | sh

# Verify installation
ollama --version
```

## Running Your First Model

```bash
# Download and run Llama 3 8B
ollama run llama3

# Or run in API mode
ollama serve
```

Then query via API:

```bash
curl http://localhost:11434/api/generate -d '{
  "model": "llama3",
  "prompt": "Explain quantum computing in simple terms"
}'
```

## Performance Tips

1. **Use GPU acceleration** — CUDA support dramatically improves speed
2. **Quantize models** — GGUF format with Q4_K_M quantization
3. **Context length** — Keep it reasonable (2048-4096 tokens)

## What's Next?

In upcoming posts, we'll explore:
- Building a local RAG system
- Fine-tuning models for specific tasks
- Creating a local AI API gateway

---

*Want to see more local AI content? Subscribe to the newsletter!*

---

# The Skill of Learning How to Learn
URL: https://saurabh.dev/blog/the-skill-of-learning-how-to-learn/
Published: 2025-01-25
Journey: Reflections
Tags: learning, learning, meta-learning, knowledge-management

We spend years in school learning subjects, but almost no time learning *how* to learn. It's like teaching someone to write without teaching them to read.

## The Core Idea

Learning is not about memorizing. It's about building mental models that you can apply to new situations.

The goal isn't to remember facts. The goal is to understand patterns.

## What Actually Works

### Active Recall

Don't re-read your notes. Close them and try to recall what you just learned.

The struggle of recalling is where the learning happens. It's counterintuitive — the harder it feels, the more you're learning.

### Spaced Repetition

Review information at increasing intervals. Your brain strengthens memories when it has to work to retrieve them.

I use Anki for this. Even for things I think I "know."

### Elaboration

Don't just learn a fact. Ask yourself: why is this true? How does it connect to what I already know? Can I explain it to someone else?

The connections are more important than the facts.

### Concrete Examples

Abstract concepts are hard to remember. Concrete examples make them stick.

For every abstract idea, find 2-3 specific examples.

## What Doesn't Work

- **Re-reading** — Feels productive, isn't
- **Highlighting** — You're just marking text, not learning it
- **Cramming** — Works for the test, not for retention
- **Multitasking** — Your brain can't actually do it

## What I'm Still Figuring Out

- How to balance depth vs. breadth
- How to maintain knowledge over time
- How to learn efficiently without losing the joy of exploration

---

*This post is part of my ongoing exploration of learning and knowledge management.*

---

# Mental Models for Better Decisions
URL: https://saurabh.dev/blog/mental-models-for-better-decisions/
Published: 2025-01-20
Journey: Decision Frameworks
Tags: thinking, mental-models, decision-making, critical-thinking

Mental models are frameworks for thinking. They help you see problems from different angles and make better decisions by giving you a structured way to approach complexity.

Here are a few I keep coming back to.

## First Principles Thinking

Break a problem down to its fundamental truths, then reason up from there.

Instead of asking "how do others solve this?", ask "what do I actually know to be true?" and build from there.

**When to use it:** When you're stuck, when conventional wisdom isn't working, when you need to innovate.

## Inversion

Instead of asking "how do I achieve X?", ask "what would guarantee failure?" then avoid those things.

Charlie Munger's favorite: "Tell me where I'm going to die, and I'll never go there."

**When to use it:** When planning, when risk assessment matters, when you're not sure what success looks like.

## Second-Order Thinking

First-order thinking: "If I do X, Y will happen."
Second-order thinking: "If I do X, Y will happen, which means Z will happen, which means..."

Most people stop at first-order. The real consequences are often in the second or third order.

**When to use it:** When decisions have long-term consequences, when you're dealing with complex systems.

## Hanlon's Razor

"Never attribute to malice that which can be adequately explained by stupidity (or negligence)."

In practice: assume good intent until proven otherwise. Most problems are coordination failures, not character flaws.

**When to use it:** When someone screws up, when you're frustrated with others, when assigning blame.

## Map Is Not the Territory

Your mental model of something is not the thing itself. Maps are useful but always incomplete.

The map is not the territory. The menu is not the meal.

**When to use it:** When you're confident about something, when your model isn't matching reality, when planning.

## What I'm Still Figuring Out

- How to know which model to apply when
- When models conflict, how to decide
- Whether thinking in models is itself a trap

---

*These are my notes, not a comprehensive guide. For deeper reading, check out "Poor Charlie's Almanack" or Farnam Street's mental models series.*

---

# Why I'm Starting This Blog
URL: https://saurabh.dev/blog/why-i-started-this-blog/
Published: 2025-01-15
Journey: Reflections
Tags: thinking, learning, meta, writing

I've started and abandoned many blogs. Usually because I tried to be too niche, too polished, or too consistent.

This time is different. I'm not trying to build a brand or become an "influencer." I just want a place to think out loud.

## What This Blog Is

A collection of ideas, learnings, and reflections on whatever I'm exploring. Sometimes technical. Sometimes philosophical. Sometimes just notes to myself that I'm making public.

## What This Blog Isn't

- Not a newsletter
- Not a personal brand
- Not a tutorial site (though there will be tutorials)
- Not a place where I pretend to have all the answers

## The Rules

1. Write when I have something to say
2. Don't optimize for engagement
3. It's okay to say "I don't know yet"
4. Topics will evolve as I do

That's it. If you're interested in the same things — tech, thinking, learning — stick around. If not, no hard feelings.

---