40
2025-09-21
Replit Agent 3
Building a micro SaaS using Replit's Agent 3. Very powerful and can debug its own code. Needs very clear and specific instructions to not waste agent usage.
Notebook
Active notes about what I'm building, breaking, and learning along the way.
39
2025-09-17
Introduction to ChatGPT Training Session
Prompt refinement: context and clarity are king. Always redact sensitive details.
38
2025-09-07
n8n + Pinecone RAG
Experimented with connecting n8n workflows to Pinecone for retrieval-augmented generation. Works but needs batching for large docs.
37
2025-09-03
Explore ZAPIer alternatives
Plan to reSearch cheaper ZAPIer alternatives like Pipedream or custom n8n cloud hosting for small business automation.
36
2025-08-30
n8n date handling
Learned you can normalize dates to ISO strings before deduping items. Prevented mismatches across feeds.
35
2025-08-22
Docubot prototype
Started the Docubot project to auto-generate docs. Early prototype outputs clean structured HTML from prompts.
34
2025-08-18
Custom embeddings with OpenAI
Experimented with different embedding models. Found ada-002 is still cost-effective but struggles with domain-specific jargon.
33
2025-08-10
Docker image size
Discovered that using python:slim cut my container size in half. Worth always starting slim and only adding needed packages.
32
2025-08-04
Custom API proxy
Built a lightweight Node.js proxy to handle API requests and secure keys. Useful for connecting frontend experiments to OpenAI.
31
2025-07-31
n8n error handling
Tried error workflows in n8n. Learned they're easier to maintain than wrapping try/catch logic around every Function node.
30
2025-07-27
Test Weaviate hybrid Search
Want to test Weaviate's hybrid sparse + dense Search. Could give better results for mixed keyword + semantic queries.
29
2025-07-23
Vector chunk size
Learned from testing that smaller embedding chunks (~300 tokens) performed better than 800+ tokens for semantic Search accuracy.
28
2025-07-15
Byteflux AI news agent
Built the Byteflux workflow to pull AI news from multiple feeds and email summaries daily. Good candidate for a portfolio demo.
27
2025-07-12
LangChain multi-step chains
Tested chaining multiple prompts together in LangChain. Learned context can get lost unless explicitly carried forward in memory.
26
2025-07-08
AWS EC2 + Amazon Linux quirks
While setting up EC2 with Amazon Linux, learned it simplifies integration with other AWS services but has older package defaults.
25
2025-07-05
Docubot RAG system
Converted my Docubot project to a RAG system.
24
2025-07-01
AI as a tool
Wrote about how AI isn't magic or scary—it's a tool. My goal is to use it for good, boosting workflows and helping others learn.
23
2025-06-27
n8n portfolio project ideas
Want to create n8n workflows to showcase in my portfolio: news agent, Slack notifier, and workflow orchestrator.
22
2025-06-23
Tagline animations
Played with gradient shimmer, glow, and heartbeat animations for my portfolio tagline. Settled on random color letter cycling.
21
2025-06-16
PHP interface for AI prompts
Created a custom PHP page that sends prompts to VaporGPT and receives structured HTML+Python back. Learned to style code blocks cleanly.
20
2025-06-12
Master prompt writing
Reflected on how writing a master prompt to describe myself saves re-explaining every time. Helps me code faster and focus on building.
19
2025-06-05
Document converter tools
Need to explore Python and PHP options for converting DOCX/PDF/RTF so I can handle uploads and serve clean text for AI training.
18
2025-06-02
n8n AI news agent
Built a daily AI-news agent in n8n: RSS → HTTP Request (custom feeds) → filter → Gmail. Learned you can normalize dates by mapping to ISO strings before deduping.
17
2025-05-30
LangChain retriever filters
Tested metadata filters in Pinecone. Found out filters must be JSON objects, not strings.
16
2025-05-25
Prompt engineering frustrations
Ran into inconsistency with zero-shot prompts. Added structured format to stabilize output.
15
2025-05-22
Automation with n8n + Gmail
Built a workflow that grabs RSS feeds, filters by keyword, and emails results to myself daily.
14
2025-05-19
JSON mode in OpenAI API
Learned that JSON mode is much cleaner for CLI tools. Parsing became trivial with schema validation.
13
2025-05-16
Dockerizing AI service
Wrapped a Python AI script into a Docker container. Biggest hurdle was setting up the slim base image.
12
2025-05-12
Developer branding
Realized that naming projects makes them memorable. A good logo adds more credibility than expected.
11
2025-05-06
Vector DB gotchas
Learned that indexing large docs works best when chunked. Optimal size ~500 tokens for embeddings.
10
2025-05-02
Zero-shot vs few-shot
Compared zero-shot and two-shot prompts. Few-shot performed more reliably for structured answers.
9
2025-04-28
Investigate Cursor editor
Check if Cursor offers better AI-assisted workflows than VS Code. Curious about context window handling and inline completions.
8
2025-04-20
Prompt tokens burn fast
Realized that careless prompts in GPT-4 eat tokens fast. Need to focus on concise instructions and context windows.
7
2025-04-20
Weaviate vs Pinecone test
Benchmarked small document sets across Weaviate and Pinecone. Weaviate was easier to self-host but Pinecone had faster queries.
6
2025-04-17
Evaluate Hugging Face Hub
Need to explore model hosting on Hugging Face Hub and see how to pull weights into local workflows.
5
2025-04-12
n8n Function node quirks
Found that Function nodes reset scope on each run. Must explicitly return objects or values won't persist.
4
2025-04-09
GitHub Actions workflow
Created a CI/CD pipeline with GitHub Actions to run tests on push. Helps keep repos clean without manual checks.
3
2025-04-05
Docs vs tutorials
Noticed I learn faster from short YouTube examples than official docs. Docs are thorough but slower to digest.
2
2025-04-05
Test pgvector
Need to check if pgvector can handle semantic Search in smaller projects without adding another vector DB service.
1
2025-04-03
OpenAI function calling
Played with function-calling to return structured JSON. Found it reliable for CLI tools but tricky when nesting functions.