Forem: HoangTheQuyen

I Built an Open-Source CLI Because My AI Kept Giving Me Fortune-Cookie Advice

HoangTheQuyen — Tue, 10 Mar 2026 10:52:45 +0000

I asked Claude to help me decide between two job offers.

It gave me a "pros and cons" list. Generic. Surface-level. The kind of advice you'd get from a random blog post.

That was the moment I realized: AI assistants are brilliant at generating text, but terrible at structured thinking.

So I built Think Better -- an open-source CLI that permanently injects decision-making frameworks into your AI's brain.

The Problem

Every time you ask an AI for help with a real decision, you get:

Pros and cons (basic)
"It depends on your situation" (useless)
No framework for HOW to think
No awareness of cognitive biases
No structured decomposition

The AI has the knowledge. It just doesn't have the methodology.

What Think Better Does

Think Better installs structured thinking frameworks directly into your AI assistant's configuration. After a one-time setup, your AI automatically:

1. Selects the right framework for your problem:

Your Question	Framework Used
"Should I take this job?"	Weighted Decision Matrix
"Why is our conversion dropping?"	Issue Tree + MECE
"What could go wrong with this launch?"	Pre-Mortem Analysis
"PostgreSQL or MongoDB?"	Evaluation Matrix

2. Detects cognitive biases in real-time:

Your AI will flag when you're falling into:

Sunk Cost Fallacy -- "You've invested time, but that shouldn't affect future decisions"
Anchoring Bias -- "Your first data point might be skewing your judgment"
Status Quo Bias -- "Are you choosing this because it's familiar?"
...plus 9 more bias detectors

3. Generates structured outputs:

Instead of paragraphs of text, you get:

Comparison matrices with weighted scores
Issue trees with MECE decomposition
Pre-mortem risk assessments
Hypothesis testing frameworks

Install in 30 seconds


bash
# Install
curl -sSL https://raw.githubusercontent.com/HoangTheQuyen/think-better/main/install.sh | bash

# Initialize for Claude
think-better init --ai claude

# Or for GitHub Copilot
think-better init --ai copilot

How I Built a Free Inflation Calculator for Vietnam's 2026 Oil Crisis

HoangTheQuyen — Mon, 09 Mar 2026 15:08:58 +0000

How I Built a Free Inflation Calculator for Vietnam's 2026 Oil Crisis
When oil prices crossed $100/barrel and gas prices in Vietnam surged overnight, I realized my personal finance tools needed to answer one urgent question: "How much is inflation actually eating my salary?"

The Problem
On March 7, 2026, Vietnam's fuel prices saw their largest single-day increase in years:

RON 95-III gasoline jumped to 27,040 VND/liter (+4,700 VND)
Diesel surged to 30,230 VND/liter (+7,200 VND)

This wasn't just about filling up your motorbike. It triggered a classic cost-push inflation domino:

Oil price ↑ → Transport costs ↑ → Raw materials ↑ → Consumer prices ↑ → Your purchasing power ↓

Vietnam's CPI for February 2026 had already rebounded to 3.35%, and forecasts pointed toward 4.5% by year-end. For someone earning 20 million VND/month (~$780 USD), that's roughly 10.8 million VND/year silently disappearing from their wallet.

I wanted to build a tool that makes this painfully visible.

The Stack
My fintech calculator ecosystem (Máy Tính Tài Chính) runs on:

Next.js 16 (Turbopack) — SSR for SEO, client-side interactivity for calculators
TypeScript — End-to-end type safety
Vanilla CSS — No Tailwind, full control over design tokens
Vercel — Edge deployment, 100/100 Lighthouse score
Building the Inflation vs Savings Calculator
The core logic is deceptively simple but financially important:

typescript

// Real purchasing power after inflation
const realValue = nominalValue / Math.pow(1 + inflationRate, years);
// Savings growth with compound interest  
const savingsGrowth = principal * Math.pow(1 + bankRate, years);
// The gap = how much you're actually losing
const purchasingPowerLoss = savingsGrowth - (savingsGrowth / Math.pow(1 + inflationRate, years));

The key insight: if your bank offers 5% interest but inflation is 4.5%, your real return is only 0.5%. On 100 million VND, that's just 500,000 VND of actual growth per year — barely enough for a dinner out.

You can try the live calculator here: Tiết Kiệm vs Lạm Phát Calculator

What I Learned About SEO for Financial Tools
Building 50+ calculators taught me that **utility pages rank differently **from blog posts:

Search intent is transactional — Users want to calculate, not read
Schema markup matters — SoftwareApplication schema gets rich snippets in Google
Internal linking is king — Each calculator links to 3-5 related tools, creating a "content mesh"
For the inflation crisis specifically, I wrote a deep-dive analysis on how rising oil prices impact Vietnamese household budgets, complete with real CPI data and a 5-step protection plan.

The Results

Lighthouse Score: 100/100 Performance, Accessibility, Best Practices, SEO
50+ financial calculators covering tax, savings, investment, debt, retirement
27 blog posts targeting Vietnamese personal finance keywords
Zero dependencies on external APIs for calculations (all client-side)

Try It Yourself
The entire ecosystem is free to use at maytinhtaichinh.com. Whether you're worried about how to protect your assets during the Iran conflict or just want to calculate your net salary after tax, there's a tool for that.

If you're building fintech tools or dealing with localized financial calculations, I'd love to hear about your approach. Drop a comment below!

I GOT TIRED OF GENERIC AI ANSEWERS. So I built a 'Brain Implant' for AI

HoangTheQuyen — Sun, 08 Mar 2026 17:30:16 +0000

If you use Claude, Copilot, or ChatGPT every day, you know exactly what I'm talking about. You ask a high-stakes engineering question like: "Should we migrate this legacy monolith to microservices or just optimize the DB?"

And 99% of the time, the AI spits out the exact same lukewarm, fence-sitting, generic "pros and cons" garbage.

The AI isn't inherently stupid—it just lacks structured mental models. It statistically predicts words instead of algorithmically solving problems.

I got tired of this. So I spent the last 3 months building Think Better — an open-source collection of "Skills" that literally forces your AI to stop hallucinating and start thinking using hard decision science.

How these "Logic Skills" work

Instead of relying on fragile system prompts, these Skills equip your AI with a local BM25 Search algorithm (via embedded Python) running over a heavily curated Knowledge Base of pure logic:

10 Strategic Decision Frameworks: (Hypothesis-Driven Trees, Pre-mortem Analysis, Weighted Matrices)
12 Cognitive Bias Detectors: It actively detects flaws in your logic. If you defend an old tech stack just because you wrote it, the AI triggers Status Quo Bias and mathematically forces you to calculate the cost of inaction.
10 Lethal Problem Solving Methods: (MECE Decomposition, Profitability Trees, Root Cause Isolation)

Upgrade your AI in exactly 30 seconds

The repository provides a Go-based installer that instantly packs and mounts these Python/CSV skills directly into your local AI workspace (.claude, .copilot, etc). It runs 100% locally. Zero API keys. Zero telemetry.

bash
git clone https://github.com/HoangTheQuyen/think-better.git
cd think-better
make build
# Install the skills into your AI assistant
./bin/think-better init --ai antigravity

Next time a production system goes down, type "Revenue dropped 20% tonight. Help me find the root cause." Instead of generic advice, your AI will trigger the Problem Solving skill, rip open a MECE Profitability Tree, and ruthlessly decompose the issue layer by layer until the bug is dead.

This is the upgrade your AI actually needs. Star the repo if you hate generic AI answers: https://github.com/HoangTheQuyen/think-better

Would love to hear feedback from builders and anyone obsessed with optimizing local AI workflows!