Twitxy Saved

rvaniaaa

Verified account This account is verified. Learn more

@rvaniaaaa

I break down the AI setups top builders actually use and show you how to copy them. Agents, second brains, cutting AI costs.
Joined September 2024
279 Following 5.9K Followers 3.9K Posts

Latest posts 5

rvaniaaa Verified account @rvaniaaaa
This is f*cking insane, the best way to print money right now is a GTA 6 clip studio that lives in your Slack

7 free repos + Fable 5.2 + Slack = an AI teammate that finds the clips, cuts them and sends them to you

why Slack? it's your control panel: you give orders in a message, the bot sends the finished clip right to your channel, you tap Approve on your phone and it posts

message → search → cut → captions → Slack review → posting queue

how to connect Slack in 10 minutes:

1. go to api.slack.com/apps and create a new app

2. turn on Socket Mode, so your bot runs from your own computer

3. give it permission to read mentions, send messages and upload files

4. install it to your workspace and copy the tokens

5. run the bot with Bolt below, and message it in your channel

Slack is the control panel, now here are the 7 repos that do the actual work ↓

-> Bolt for Python (by slackapi) github.com/slackapi/bolt-…
Slack's official framework, your bot reads your messages and posts clips back in the channel

-> browser-use (by browser-use) github.com/browser-use/br…
lets an agent search the web for clips on its own, then sends you the list

-> Whisper (by openai) github.com/openai/whisper
turns speech into text, so the bot finds the best moments by words

-> Ultralytics (by ultralytics) github.com/ultralytics/ul…
detects cars, people and fights in the frame, so the bot picks highlights for you

-> FFmpeg (by FFmpeg) github.com/FFmpeg/FFmpeg
cuts, encodes and combines every clip from one command

-> Remotion (by remotion-dev) github.com/remotion-dev/r…
renders captions as code, so every clip gets your style automatically

-> Postiz (by gitroomhq) github.com/gitroomhq/post…
schedules posts across social accounts, approved clips publish on time

Let Fable 5.2 write the glue code between the tools. Start with one bot that answers one message, then add a step at a time
nothing goes live until you approve it in Slack

prompt to build the whole bot below ↓
rvaniaaa Verified account @rvaniaaaa Nov 11
View quoted post on X
1 replies 1 reposts 12 likes 477 views View on X
rvaniaaa Verified account @rvaniaaaa
THIS IS F*CKING INSANE, ANTHROPIC PUBLISHED ITS OWN COURSES ON GITHUB FOR FREE, AND 11 MORE REPOS TAKE YOU FROM ZERO TO AI ENGINEER

12 repos, 12 weeks, one plan: Python → prompting → LLMs → fine-tuning → RAG → agents → evals → production

Each week gives you something to build, test, and explain

01 Python foundations: Learn Python
Syntax, data structures, and the coding skills everything else depends on.
github.com/trekhleb/learn…

02 Prompting: Anthropic
An interactive tutorial that teaches you to write prompts that work on the first try.
github.com/anthropics/pro…

03 Claude API: Anthropic
Courses on the API, prompt engineering, tool use, and evaluations.
github.com/anthropics/cou…

04 LLMs from scratch: Sebastian Raschka
Code a GPT style model step by step and understand how it really works.
github.com/rasbt/LLMs-fro…

05 The full LLM roadmap: Maxime Labonne
Fundamentals, the science, and the engineering of large language models in one place.
github.com/mlabonne/llm-c…

06 Train your own GPT: Andrej Karpathy
A small, fast codebase for training and fine-tuning GPT models.
github.com/karpathy/nanoG…

07 Fine-tuning: Hugging Face
Adapt a pretrained model to your own task.
github.com/huggingface/sm…

08 RAG: Nir Diamant
Techniques for building assistants that answer from your own documents.
github.com/NirDiamant/RAG…

09 AI agents: Hugging Face
Build systems that use tools and carry out tasks through agent workflows.
github.com/huggingface/ag…

10 Claude recipes: Anthropic
Working examples for tool use, retrieval, and agent patterns you can copy.
github.com/anthropics/cla…

11 Evaluation: Confident AI
Test and score your LLM apps before your users find the bugs.
github.com/confident-ai/d…

12 AI engineering: Chip Huyen
The code and notes behind the book on building AI applications in production.
github.com/chiphuyen/aie-…

Follow one step per week. Finish a small project before moving on

Your target portfolio: a model-training project, a RAG assistant with citations, and a tool-using agent with evaluations

Plan for 10 to 15 hours a week with basic Python knowledge. The repos are free; APIs and compute may cost extra

Pick the first repo you need. Build something worth putting your name on

Save this ↓
0:30
rvaniaaa Verified account @rvaniaaaa Nov 11
View quoted post on X
1 replies 4 reposts 32 likes 1.3K views View on X
rvaniaaa Verified account @rvaniaaaa
This is f*cking crazy, Anthropic, OpenAI, Google and Microsoft open sourced the tools they build their own agents with

Any one of these 7 repos is enough to build your own agent this week

-> claude agent sdk python (by anthropics) github.com/anthropics/cla…
the agent loop behind Claude Code, with file, shell and web tools, subagents and tool permissions built in

-> openai agents python (by openai) github.com/openai/openai-…
a minimal agent SDK with handoffs, tracing on by default and sandbox agents

-> adk python (by google) github.com/google/adk-pyt…
LLM driven agents plus graph workflows, deploy to Google Cloud or any container

-> agent framework (by microsoft) github.com/microsoft/agen…
the successor to AutoGen and Semantic Kernel, with graph workflows and a harness agent for long tasks

-> harness sdk (by strands-agents) github.com/strands-agents…
an SDK that runs in your own process with no hosted control plane, plus MCP and multi-agent patterns

-> langgraph (by langchain-ai) github.com/langchain-ai/l…
stateful agent workflows with checkpointing, crash recovery and human pauses

-> markitdown (by microsoft) github.com/microsoft/mark…
turns PDFs, Office files and web pages into Markdown, the first step of any second brain

the labs gave away the toolkits
pick one and build your agent
0:30
rvaniaaa Verified account @rvaniaaaa Nov 11
View quoted post on X
8 replies 7 reposts 41 likes 1.6K views View on X
rvaniaaa Verified account @rvaniaaaa
This is f*cking smart, the CEO of Chroma open sourced the research on why AI gets dumber as context grows, and these 7 free repos show you how to fix it

-> context rot (by chroma-core) github.com/chroma-core/co…
Chroma's open research code that tests how longer inputs make models worse, run it on any model

-> chroma (by chroma-core) github.com/chroma-core/ch…
the open source search database behind it, with regex search and index forking

-> context 1 data gen (by chroma-core) github.com/chroma-core/co…
the data pipeline behind Context-1, Chroma's open 20B search agent that finds the right documents for a bigger model

-> context engineering intro (by coleam00) github.com/coleam00/conte…
a template that gives Claude Code structured project context with CLAUDE.md, INITIAL.md and examples

-> context engineering (by davidkimai) github.com/davidkimai/Con…
a first principles handbook for designing everything a model sees, not just the prompt

-> awesome context engineering (by Meirtz) github.com/Meirtz/Awesome…
a survey of papers and tools on RAG, memory, agent protocols, coding agents and observability

-> advanced context engineering for coding agents (by humanlayer) github.com/humanlayer/adv…
HumanLayer's repo with Dex Horthy's essays on getting coding agents to work

more context is not smarter
the right context is
0:30
rvaniaaa Verified account @rvaniaaaa Sep 20
View quoted post on X