Notes from the workbench.
Practical notes on AI agents, local models, and the systems behind them. Browse 57 posts, search the transcripts, or watch on YouTube.
Pick a thread
AI coding agents
Understand how agents find context, call tools, and work within the boundaries of a real repository.
4 handpicked posts02Local AI
Models, memory, and practical tradeoffs when the hardware running your AI is your own.
4 handpicked posts03Data & backend systems
Trace pipelines, inspect their data, and move slow work out of the request path.
3 handpicked posts57 posts · 55 videos · 2 articles
7 min readVideo
An OpenAI Agent Escaped Its Sandbox—and AI Had Its Wildest Week
A benchmark agent found a zero-day in its own package proxy, reached the internet, and pulled the answers from Hugging Face. Long-horizon autonomy moves the unit of safety from the single tool call to the whole trajectory.
Watch on YouTube7 min readVideo
How I Built a Local AI Talking Head That Keeps the Real Person
Instead of regenerating the person, the pipeline keeps the real recording and synthesizes only the pixels inside a mouth mask. SyncNet holds the normal track at a zero-frame offset, and a deliberate 400 ms shift moves it by 10 frames — the evaluator is tracking timing, not faces.
Watch on YouTube6 min readVideo
Why AI Agents Fail After the Demo—and What CARS24 Gets Right
CARS24 reports more than a million agent conversation minutes a month while most agent demos collapse the moment the task leaves a clean prompt. The difference is a five-part contract around the model: accepted outcome, bounded permissions, evidence-based routing, trajectory evaluation, and a fallback.
Watch on YouTube5 min readVideo
Remotion: The Part Most Tutorials Miss
Remotion is the camera and the stage, not the director. The polish comes from four visual rules, three reusable templates, and a review loop that judges rendered frames at phone size instead of the source code.
Watch on YouTube8 min readVideo
Why AI Coding Agents Ship Unreviewed Work
A review approves artifact A, a retry drops artifact B at the same path, and the saved approval still passes because it was never bound to the bytes. A read-only verifier recomputes the digest at the action boundary and refuses the replacement.
Watch on YouTube1 min readVideo
109 Focused Tests Passed. The Final Video Still Wasn't Proven.
A green test suite did not prove that the final MP4 still matched the approved package. The repair binds the current video path and SHA-256 digest to the validation report.
Watch on YouTube1 min readVideo
The Identity Check This AI Video Pipeline Was Missing
The publish gate now requires the reviewed package, story, current video path, and current bytes to agree, stopping stale state or the wrong artifact before release.
Watch on YouTube1 min readVideo
The Prompt Constraint That Kept This GPT-5.6 Sol Refactor Bounded
One behavior, three integration boundaries, 12 application files, and no merge authority: the constraint that kept a real GPT-5.6 Sol refactor narrow enough to review.
Watch on YouTube1 min readVideo
I Gave GPT-5.6 Sol a Real Repo Refactor
A bounded GPT-5.6 Sol refactor inside a real production video system, followed by an independent review that found the missing artifact-provenance guarantee.
Watch on YouTube1 min readVideo
Your Local AI Tools Should Train From Their Logs
Local assistants already record failures, retries, repairs, and corrections. With training events, redaction, quality gates, and evals, those logs can become a disciplined learning loop.
Watch on YouTube5 min readVideo
How Fish-S2 Plus ClearVoice Made The Voice Feel Real
An AI voice pipeline is only useful if it can reject bad takes before they ship. Fish-S2 Plus generates the narration, ClearVoice cleans each slide after the mix, and NIQA plus Audio Flamingo score artifacts, speaker match, and expression so cleanup improves rough audio without sanding the voice flat.
Watch on YouTube12 min readVideo
Run AI On Your Laptop — Zero Cloud Bills (Local Small Models)
Local small models are the right default for narrow, high-volume AI work: classification, extraction, routing, and rewrites that should not be billed forever. Run them through Ollama, MLX, or vLLM, fine-tune with LoRA when prompting drifts, and escalate only the hard cases to the cloud.
Watch on YouTube16 min readVideo
How AI Generates Text vs Images
Text and image generation look like the same trick from the outside, but under the hood they are two completely different machines: a language model that writes one token at a time like a typewriter, and an image model that sculpts a picture out of noise. Different math, different training, different failure modes.
Watch on YouTube10 min readVideo
Headroom: Your AI Agent Is Wasting Money On Tokens
The most expensive part of an AI agent is usually not the answer - it is everything the agent reads on the way there. Headroom compresses that context before the request leaves your machine, keeps the original nearby, and lets the model ask for the full detail only when it needs it.
Watch on YouTube6 min readVideo
Can Self-Hosted Models Do Real Agentic Work?
The honest, fast-changing answer to whether an open model running on hardware you own can do genuinely useful agentic work inside a real codebase - what you trade away by renting a cloud frontier model, and what self-hosting actually buys back.
Watch on YouTube13 min readVideo
What Actually Fits on 128 GB (Quantization Explained)
Quantization is the single most important lever for fitting large models onto hardware you actually own. What a number really is inside a model, how quantization shrinks it, and what genuinely fits in 128 GB of unified memory.
Watch on YouTube16 min readVideo
The Local AI Loop the Cloud Can't Afford
A small box on your desk can now run a whole family of open models at once - and do the one move the cloud quietly will not let you afford: let the models loop on their own outputs. Why raw speed is the wrong scoreboard for local AI.
Watch on YouTube6 min readVideo
NVIDIA Nemotron 3 Ultra Changes AI Agents
NVIDIA's Nemotron 3 Ultra is built as the flagship model for long-running agents - planning across steps, calling tools, holding context, and recovering when the first attempt fails. Where Ultra belongs in an engineering stack, and when a leaner focused model is the better tool.
Watch on YouTube7 min readVideo
I Turned My Gaming Rig + DGX Spark Into One AI Cluster
Joining a gaming laptop with an RTX 4090 and an NVIDIA DGX Spark into a single Kubernetes cluster so they work as one pool of compute - and a from-the-ground-up Kubernetes lesson using that real cluster as the running example.
Watch on YouTube14 min readVideo
Blender Makes Better Motion Graphics Than You Think
The slick animations in explainer videos - sliding titles, sweeping arrows, counting numbers - don't need expensive software or a paid studio. A huge number are made in Blender, which is free. From clean flat graphics to full three-dimensional showpieces for your channel.
Watch on YouTube12 min readVideo
Stop Making Users Wait: Async Queues Explained
Async processing as a handoff: Django accepts the request, Redis holds the job, and a Celery worker finishes the slow work after the response is already back. Queues, workers, backlogs, retries, and when not to bother.
Watch on YouTube15 min readVideo
pgvector: Vector Search Inside Postgres (No Extra DB)
Why pgvector lets Postgres do vector search itself - a vector column type, distance operators, and IVFFlat vs HNSW indexes - so most projects never need a separate vector database.
Watch on YouTube3 min readVideo
LangSmith: Close the Loop Between Shipped and Working
LangSmith's agent engine closes the gap between shipped and working: it clusters production failures into named issues, traces each back to the commit that introduced it, and drafts a fix that waits for your approval.
Watch on YouTube7 min readVideo
Tavily For Beginners: Give Your AI Real-Time Web Search
Tavily gives a language model real-time web search in four lines - structured results with a synthesized answer field, search-depth and time-range knobs, and a first-party LangChain tool.
Watch on YouTube10 min readVideo
Running A RAG Pipeline On The Pentagon UFO Files — Real Cypher, Real Citations
Episode 2: open the actual repo and run all six stages on 115 declassified PDFs — Chroma retriever returns cited answers and a FalkorDB graph agent writes Cypher from plain English.
Watch on YouTube9 min readVideo
Teach AI To Read UFO Files (LangGraph + RAG)
Episode 1: the architecture of a six-stage RAG pipeline plus a GraphRAG layer in LangChain + LangGraph, built to read 4 GB of declassified UFO files on a single workstation.
Watch on YouTube9 min readVideo
LangChain Explained for Engineers: How the Runnable Interface Works
Once the Runnable interface clicks, the rest of LangChain — LCEL, LangGraph, tools, retrievers, structured output — becomes predictable instead of surprising.
Watch on YouTube8 min readVideo
LangGraph Local-First: StateGraph, Reducers, and Custom Chat Models
LangGraph's StateGraph, reducers, conditional edges, and checkpointer run identically against local models - no hosted LLM or API key required.
Watch on YouTube15 min readVideo
20 Essential 3D Printing Terms: Fillet, Chamfer, Manifold
Twenty precise 3D printing terms - fillet, chamfer, manifold, draft angle, clearance fit - that let you command an AI to design parts that actually print on the first try.
Watch on YouTube16 min readVideo
Blender MCP: Export 3MF and Automate 3D Prints via AI
Wiring Blender's MCP server and 3MF export into Claude Code or Codex so AI agents can drive a full 3D print workflow end-to-end.
Watch on YouTube15 min readVideo
LangChain Framework Explained: LCEL, LangGraph & RAG
A complete tour of LangChain's five packages - LCEL, LangGraph, retrievers, tools, structured output - assembled into a working RAG application.
Watch on YouTube21 min readVideo
CADQuery and AI: Automating Parametric 3D Printing from Text
How AI plus CADQuery turns text descriptions into parametric CAD scripts and STL files - and where the workflow quietly breaks.
Watch on YouTube16 min readVideo
Web Scraping in the AI Era: From BeautifulSoup to LLM Extraction
Every web page is structured data wearing an HTML costume. The four moves under the hood, the three-layer stack, and what AI changed about all of it.
Watch on YouTube13 min readVideo
How to Get Cited by AI Answer Engines
What actually makes a page show up inside ChatGPT, Perplexity, and Google AI Overviews — and how to write for it.
Watch on YouTube18 min readVideo
How Local AI Actually Remembers: Inside The Agent Log
Every local coding agent writes a structured log. Resume, replay, and audit are all downstream of that one file.
Watch on YouTube13 min readVideo
How Pi Mono Actually Works in Your App
Walking through the Pi Mono shared agent stack from inside a real app, not as an abstract architecture diagram.
Watch on YouTube6 min readVideo
Fish Speech S2 Pro: Open-Source Voice AI That Beats Closed Models
Hands-on with Fish Speech S2 Pro — what it sounds like, what it costs to run locally, and where it actually wins.
Watch on YouTube11 min readVideo
How Pi Mono Actually Works: The Shared Agent Stack Behind Pi
The architecture under Pi Mono — one harness, many agents — and why that shape keeps showing up in local AI tools.
Watch on YouTube7 min readVideo
I Tested Claude Code Tools — Here's What Surprised Me
The lesser-known Claude Code tools, what they really do, and which ones changed how I move through a repo.
Watch on YouTube2 min readVideo
How Repo Rules, MCP, and Approvals Control AI Coding Agents
The three control surfaces that decide what an agent can read, run, and change inside your repo.
Watch on YouTube5 min readVideo
Why Cursor Picks the Wrong Files In Large Repos (And How to Fix It)
What Cursor's retrieval is really doing in a large monorepo, why it drifts, and the small repo changes that pull it back on track.
Watch on YouTube3 min readVideo
OpenAI Codex Is Now a Multi-Agent Command Center
Codex grew up: cloud tasks, sub-agents, and approvals stitched into one workflow you can actually drive from the CLI.
Watch on YouTube8 min readVideo
Codex Explained for Engineers: How the Coding Agent Actually Works
A plain-English walkthrough of the Codex loop: instructions in, tools and reads, edits out, approvals around it.
Watch on YouTube7 min readVideo
Cursor Explained for Engineers: How the AI IDE Actually Works
What Cursor is doing under the hood — context, retrieval, edits, and the trade-offs that come with the IDE shape.
Watch on YouTube3 min readVideo
How Codex Actually Finds the Right Files
AGENTS.md, skills, MCP, search, and compaction — the layers Codex uses to narrow a big repo down to a working set.
Watch on YouTube8 min readVideo
How MCP Actually Works: The Capability Bus Behind Codex, Claude Code, and Gemini CLI
MCP, demystified. Why it's a capability bus, not a plugin store, and what that changes about how agents reach external tools.
Watch on YouTube8 min readVideo
How AI Coding CLIs Actually Work: Codex, Claude Code, Gemini CLI
Side-by-side teardown of three terminal coding agents — the loop, the tools, and the parts they each get right.
Watch on YouTube8 min readVideo
OpenClaw Explained: How the Local AI Control Plane Works
Inside OpenClaw — a local control plane for routing AI work between models, tools, and approvals on your machine.
Watch on YouTube4 min readVideo
Why Terminal AI Is The Fastest Workflow For Developers
The case for keeping AI coding in the terminal: tighter loops, smaller context, and tools you already trust.
Watch on YouTube9 min readVideo
Bruno for Local AI APIs: Faster, Reviewable Workflows
Using Bruno to drive local AI APIs — file-based requests, OpenAPI sync, and reviewable diffs instead of throwaway curl.
Watch on YouTube7 min readVideo
Build a Local AI Coding App in Neovim
Wiring a real local AI coding workflow into Neovim — the editor stays the editor, the agent stays in the loop.
Watch on YouTube5 min readVideo
ComfyUI Explained for Engineers: How to Build Local AI Apps
ComfyUI from a software engineer's angle — nodes as functions, graphs as programs, and where the real product lives.
Watch on YouTube8 min readVideo
Azure Data Factory with Azure CLI - Part 2: Inspecting Pipelines & JSON
Deeper dive into Azure Data Factory pipeline JSON: activities, parameters, variables, and datasets via the Azure CLI.
Watch on YouTube3 min readVideo
Azure Data Factory with Azure CLI - Part 1: Discovering Pipelines & Datasets
Read-only Azure CLI walkthrough for Azure Data Factory: list factories, pipelines, datasets, and learn how to inspect pipeline definitions.
Watch on YouTube3 min readVideo
How to Use Codex & Apply Git Patches
Quick tutorial on how to use Codex Web with GitHub, create pull requests, and apply/copy Git patches. Includes transcript and example commands.
Watch on YouTube2 min readArticle
Explaining @PostConstruct & Factory Pattern in Java Spring
Quick example of how @PostConstruct works in Spring and how to pair it with a simple factory pattern.
2 min readArticle
Run MySQL 5.7 on Apple Silicon (M1/M2) with Docker
How to run MySQL 5.7 on an M1/M2 Mac by forcing an x86_64 platform in docker-compose. Includes the exact compose file, quick steps, and a short video.
/ Let's build
Open to opportunities.
Interested in software engineering roles focused on data integration, developer tooling, or cross-platform product work — especially where data accuracy and reliability are non-negotiable.