Newsletter Archive

Codon – Mix Compiled Speed with Python Libraries

Leave a Comment / Newsletter Archive / Khuyen Tran

Grab your coffee. Here are this week’s highlights.

📅 Today’s Picks

Codon – Mix Compiled Speed with Python Libraries

Problem
Compiled languages can make Python workloads faster, but they often come with a tradeoff.
You need to learn new syntax and give up easy access to libraries like pandas, scikit-learn, and matplotlib.
Solution
Codon lets you keep using Python libraries while speeding up the parts that benefit from compilation.
Key benefits:

Use the @python decorator to run Python code inside Codon
Keep using libraries like pandas and matplotlib from Codon code
Mix compiled Codon functions with regular Python code
Convert values between Codon and Python when needed

SenseLab – Switch Coding Agents Without Re-Explaining Context

Problem
Most coding agents keep their working memory inside one chat session, one model, or one IDE.
Switching tools means explaining the task again, and the new agent may not know which tradeoffs were made or why certain decisions worked.
Solution
SenseLab keeps the working context available across sessions, models, and coding tools.
The next agent can pick up from the previous agent’s saved context instead of starting over.
Key capabilities:

Shared memory across Codex, Claude, and other MCP clients
Multi-agent coordination, so agents can see each other’s work before they overlap
Continuous learning from past outcomes, so agents know what worked before
Decision lineage helps you debug what went wrong with an agent

️📦 Link to SenseLab: https://www.sense-lab.ai/

Stay Current with CodeCut

Actionable Python tips, curated for busy data pros. Skim in under 2 minutes, three times a week.

.codecut-subscribe-form .codecut-input {
background: #2F2D2E !important;
border: 1px solid #72BEFA !important;
color: #FFFFFF !important;
}
.codecut-subscribe-form .codecut-input::placeholder {
color: #999999 !important;
}
.codecut-subscribe-form .codecut-subscribe-btn {
background: #72BEFA !important;
color: #2F2D2E !important;
}
.codecut-subscribe-form .codecut-subscribe-btn:hover {
background: #5aa8e8 !important;
}

.codecut-subscribe-form {
max-width: 650px;
display: flex;
flex-direction: column;
gap: 8px;
}
.codecut-input {
-webkit-appearance: none;
-moz-appearance: none;
appearance: none;
background: #FFFFFF;
border-radius: 8px !important;
padding: 8px 12px;
font-family: ‘Comfortaa’, sans-serif !important;
font-size: 14px !important;
color: #333333;
border: none !important;
outline: none;
width: 100%;
box-sizing: border-box;
}
input[type=”email”].codecut-input {
border-radius: 8px !important;
}
.codecut-input::placeholder {
color: #666666;
}
.codecut-email-row {
display: flex;
align-items: stretch;
height: 36px;
gap: 8px;
}
.codecut-email-row .codecut-input {
flex: 1;
}
.codecut-subscribe-btn {
background: #72BEFA;
color: #2F2D2E;
border: none;
border-radius: 8px;
padding: 8px 14px;
font-family: ‘Comfortaa’, sans-serif;
font-size: 14px;
font-weight: 500;
cursor: pointer;
text-decoration: none;
display: flex;
align-items: center;
justify-content: center;
transition: background 0.3s ease;
}
.codecut-subscribe-btn:hover {
background: #5aa8e8;
}
.codecut-subscribe-btn:disabled {
background: #999;
cursor: not-allowed;
}
.codecut-message {
font-family: ‘Comfortaa’, sans-serif;
font-size: 12px;
padding: 8px;
border-radius: 6px;
display: none;
}
.codecut-message.success {
background: #d4edda;
color: #155724;
display: block;
}
@media (max-width: 480px) {
.codecut-email-row {
flex-direction: column;
height: auto;
gap: 8px;
}
.codecut-input {
border-radius: 8px;
height: 36px;
}
.codecut-subscribe-btn {
width: 100%;
text-align: center;
border-radius: 8px;
height: 36px;
}
}

Codon – Mix Compiled Speed with Python Libraries Read More »

llama.cpp – Run local LLMs on any hardware with one binary

Leave a Comment / Newsletter Archive / Khuyen Tran

Grab your coffee. Here are this week’s highlights.

📅 Today’s Picks

Headroom – Cut agent tokens without touching your prompts

Problem
AI agents can waste tokens on context they do not fully need.
Long logs, full files, API responses, and retrieved documents often contain repeated structure that slows the model down.
Solution
Headroom compresses what your agent reads before it reaches the model, helping reduce token usage and cost.
Key benefits:

Compresses JSON, code, logs, diffs, and plain text
Keeps original content available when the agent needs more detail
Works with Claude Code, Codex, Cursor, or directly as a Python library

llama.cpp – Run local LLMs on any hardware with one binary

Problem
Running an open LLM locally usually means a heavy Python stack, CUDA drivers, and several gigabytes of dependencies.
This adds friction everywhere: installation takes longer, dependencies break more easily, and the same model may behave differently across hardware.
Solution
llama.cpp makes local inference more portable by replacing the heavy Python/CUDA setup with a C++ runtime that supports GGUF models.
Key benefits:

No heavy Python environment required
Works on CPUs, Mac GPUs, NVIDIA GPUs, and AMD GPUs
Loads GGUF-quantized models designed for lower memory use
Direct Hugging Face model loading
Built-in tools for chatting, serving, benchmarking, and quantizing models

Stay Current with CodeCut

Actionable Python tips, curated for busy data pros. Skim in under 2 minutes, three times a week.

llama.cpp – Run local LLMs on any hardware with one binary Read More »

CloakBrowser – Make Browser Automation Look Real

Leave a Comment / Newsletter Archive / Khuyen Tran

Grab your coffee. Here are this week’s highlights.

📅 Today’s Picks

CloakBrowser – Make Browser Automation Look Real

Problem
Playwright is a browser automation framework that lets you control Chrome, Firefox, and Safari-like browsers with code.
But the browser Playwright launches does not always look like a normal user’s browser. Sites can detect those differences and respond with CAPTCHAs, blocked access, broken login flows, or misleading test results.
Solution
CloakBrowser replaces the default browser used by Playwright or Puppeteer with a more realistic Chrome-like browser.
It helps the browser behave more like a real user session, reducing CAPTCHAs, blocked pages, and broken login flows in automated workflows.

uv – Apply Dependency Overrides Without Side Effects

Problem
A global override can affect every package that uses the same dependency, creating new conflicts in parts of the project that were already working.
Solution
uv 0.11.25 introduced scoped dependency overrides.
You can apply the override only to the package that needs it instead of changing dependency resolution for the whole project.
This keeps temporary compatibility fixes from affecting unrelated packages.

📖 Read the docs

Stay Current with CodeCut

Actionable Python tips, curated for busy data pros. Skim in under 2 minutes, three times a week.

CloakBrowser – Make Browser Automation Look Real Read More »

CLI-Anything – Turn Any Desktop App into an Agent-Ready CLI

Leave a Comment / Newsletter Archive / Khuyen Tran

Grab your coffee. Here are this week’s highlights.

📅 Today’s Picks

Caveman – Make AI Coding Replies Cheaper

Problem
Extra framing and repeated context make AI replies longer, more expensive, and harder to scan.
Solution
Caveman makes agent replies shorter while preserving code, commands, paths, errors, and technical meaning.
You get the same technical answer with fewer tokens spent on filler.

CLI-Anything – Turn Any Desktop App into an Agent-Ready CLI

Problem
A common workaround is to automate the visible interface: click buttons, wait for screens to load, and infer progress from screenshots.
But the agent cannot easily see the underlying state, repeat the same steps reliably, or recover when the interface changes.
Solution
CLI-Anything does the upfront work of building a command-line layer for a desktop app.
It looks at how the app works, creates commands for common workflows, and turns them into an installable CLI.
Once that CLI exists, agents can operate the app through commands instead of screenshots and clicks.

Stay Current with CodeCut

Actionable Python tips, curated for busy data pros. Skim in under 2 minutes, three times a week.

CLI-Anything – Turn Any Desktop App into an Agent-Ready CLI Read More »

Vulture – Scan Python Projects for Dead Code

Leave a Comment / Newsletter Archive / Khuyen Tran

Grab your coffee. Here are this week’s highlights.

📅 Today’s Picks

Vulture – Scan Python Projects for Dead Code

Problem
Dead code creates maintenance cost without adding value to the current workflow. However, asking an LLM to find dead code can be expensive and inconsistent.
It only sees the context you provide, may invent usage patterns that do not exist, and can return different cleanup suggestions for the same project.
Solution
Vulture scans your code locally with static analysis, so the same input produces the same findings without token cost.
You can use Vulture for the repeatable dead-code scan, then bring in an LLM for the refactor.

Repomix – Turn GitHub Repos into AI-Ready Context

Problem
You might want an AI assistant to scan a repo so it can understand how the files work together.
But without controlled context, it may read generated files, old tests, docs, or unrelated modules before reaching the code that matters.
Solution
Repomix turns the repo into a prepared context file with filtering, compression, and token counts.
You decide what gets packed before sending it to the AI, so the prompt stays focused and cheaper.

Stay Current with CodeCut

Actionable Python tips, curated for busy data pros. Skim in under 2 minutes, three times a week.

Vulture – Scan Python Projects for Dead Code Read More »

Understand Anything – Turn Codebases into Knowledge Graphs

Leave a Comment / Newsletter Archive / Khuyen Tran

Grab your coffee. Here are this week’s highlights.

📅 Today’s Picks

gws – Give coding agents Drive access with one CLI

Problem
Claude, Codex, and other coding agents are comfortable with shell commands.
But asking them to automate Google Drive often means making them handle OAuth, API request bodies, pagination, uploads, and response parsing before they can do the actual task.
Solution
gws makes Google Workspace easier for agents to use by wrapping Drive tasks in predictable CLI commands.
Instead of building a custom Drive client, an agent can use gws to:

Inspect Drive folders and files
Upload generated files or reports
Handle pagination when Drive has many files
Read structured JSON output for downstream steps

Understand Anything – Turn Codebases into Knowledge Graphs

Problem
Have you ever opened a large codebase and wondered where to start? A common shortcut is to ask an LLM to explain files one by one.
But it is still hard to visualize how the pieces connect unless you spend time building your own graph, diagrams, and navigation interface.
Solution
Instead of manually building diagrams, Understand Anything creates a visual map of how your codebase fits together with a single slash command: /understand.
After running /understand, you can:

Explore files, functions, classes, and dependencies as an interactive knowledge graph
Find code by name or meaning with fuzzy and semantic search
Ask natural-language questions with /understand-chat
Generate an onboarding guide for new contributors

Stay Current with CodeCut

Actionable Python tips, curated for busy data pros. Skim in under 2 minutes, three times a week.

Understand Anything – Turn Codebases into Knowledge Graphs Read More »

PySpark – Write Reusable SQL Queries Without Hardcoding Values

Leave a Comment / Newsletter Archive / Khuyen Tran

Grab your coffee. Here are this week’s highlights.

📅 Today’s Picks

Git Revert vs Reset – Choose the Right Undo Method

Problem
Have you made a commit mistake? Every data scientist faces this situation, but choosing the wrong recovery method can create bigger problems than the original error. The key is understanding when to use git revert versus git reset.
Solution
The decision depends on one critical question: “Has this commit been shared?”
If YES, use git revert which adds new commits that reverse previous changes while preserving history for audit trails.
If NO, use git reset which removes commits entirely and rewrites the timeline for clean, readable history.
Both are powerful tools when used correctly.

📖 Dive deeper in Production Ready Data Science

PySpark – Write Reusable SQL Queries Without Hardcoding Values

Problem
Hardcoding values into PySpark SQL queries leads to messy, repetitive code that’s hard to maintain.
Solution
Instead, define parameterized SQL templates in Apache Spark. These templates let you insert dynamic variables into your queries, making them reusable, testable, and easier to manage across your codebase.

📖 View Full Article

Stay Current with CodeCut

Actionable Python tips, curated for busy data pros. Skim in under 2 minutes, three times a week.

PySpark – Write Reusable SQL Queries Without Hardcoding Values Read More »

PydanticAI – Control OpenAI Data Retention with openai_store

Leave a Comment / Newsletter Archive / Khuyen Tran

Grab your coffee. Here are this week’s highlights.

📅 Today’s Picks

PydanticAI: Control OpenAI Data Retention with openai_store

Problem
By default, OpenAI may retain your API request data for internal review and model improvement. For healthcare, finance, and legal applications, this default creates compliance risks you can’t afford.
Solution
PydanticAI provides the openai_store setting to explicitly disable data retention in one line.

📖 View Full Article

DuckDB – read an entire CSV folder in one line

Problem
Working with multiple CSV files often means writing extra code to load, combine, and process each file.
Solution
DuckDB makes this simpler by letting you query multiple CSV files directly with SQL and pattern matching, so you can analyze them in one efficient query.

📖 View Full Article

Stay Current with CodeCut

Actionable Python tips, curated for busy data pros. Skim in under 2 minutes, three times a week.

PydanticAI – Control OpenAI Data Retention with openai_store Read More »

dotenvx – Separate Dev, Staging, and Prod Configs Safely

Leave a Comment / Newsletter Archive / Khuyen Tran

Grab your coffee. Here are this week’s highlights.

📅 Today’s Picks

dotenvx – Separate Dev, Staging, and Prod Configs Safely

Problem
A single .env file is fine for a small project.
Once you add dev, staging, production, and CI, those settings need to be separated because each environment connects to different resources and should not share the same secrets.
Solution
dotenvx makes each environment explicit: load .env.production, .env.staging, or .env.ci directly with -f.
Because each file can be encrypted, you can commit the config safely without sharing the private keys across environments.

Ideogram 4 – Run Open-Weight Image Generation Locally

Problem
Most image generation tools are hosted behind an API. That works for quick experiments, but it limits control over cost, infrastructure, privacy, and customization.
Solution
Ideogram 4 changes that with an open-weight release. You can download the 9.3B parameter model files and run them on your own machine or server.
Key features:

Strong text rendering inside images, with a 0.97 score on an English text-accuracy benchmark
Structured JSON prompts instead of only free-form sentences
Control over subject, style, lighting, typography, and color palettes by hex code
Bounding-box coordinates for placing visual elements exactly where you want them

Stay Current with CodeCut

Actionable Python tips, curated for busy data pros. Skim in under 2 minutes, three times a week.

dotenvx – Separate Dev, Staging, and Prod Configs Safely Read More »

PP-OCRv6 – Extract Text from Receipts with Small OCR Models

Leave a Comment / Newsletter Archive / Khuyen Tran

Grab your coffee. Here are this week’s highlights.

📅 Today’s Picks

LM Studio – Pick Local Models That Fit Your Machine

Problem
Choosing a local LLM is not just about picking the smartest model.
You also need to know whether your machine can actually run it.
Model size, quantization, GPU memory, and system RAM all affect whether the model will load successfully.
Solution
LM Studio makes local model selection easier by showing the details that affect whether a model will run well.
What LM Studio shows:

Download options: choose among model variants by comparing format, quantization, and download size.
Fit signal: see whether the model is likely to fit your machine before downloading.
README: review model-specific instructions and benchmarks from the model page.

📖 View LM Studio

PP-OCRv6 – Extract Text from Receipts with Small OCR Models

Problem
When extracting text from messy documents, many teams now reach for large vision-language models.
But if you only need to find and read text, a smaller OCR-specific model may be faster and cheaper to run.
Solution
PP-OCRv6 is designed for reading text from images, rather than broad vision-language reasoning, so it can focus on text extraction efficiently.
It delivers strong text detection and recognition while staying much smaller than billion-parameter VLMs.
With tiny, small, and medium variants, you can start lightweight and switch to a larger model only when the extracted text needs better accuracy.

🧪 Run code

Stay Current with CodeCut

Actionable Python tips, curated for busy data pros. Skim in under 2 minutes, three times a week.

PP-OCRv6 – Extract Text from Receipts with Small OCR Models Read More »

Get in touch

Join the Newsletter

Follow Us on Social Media

Newsletter Archive

Work with Khuyen Tran

Work with Khuyen Tran