Agentic_AI

This repo is my learning playground for agentic AI—basically, figuring out what an “AI engineer” actually builds by shipping lots of small, runnable experiments.

It’s not meant to be a polished framework. The point is to learn by building and keeping the projects small enough that I can iterate fast.

What I’m practicing here:

• the agent loop: observe → decide → act
• tool use (LLMs calling functions/tools)
• prompt + system instruction design
• real-world tradeoffs like cost, latency, and reliability

What’s inside

Each folder is its own mini-project you can run independently:

• browser-agent/
  A browser automation agent (Playwright + Gemini) that can navigate real websites. Focus: tool loop, extracting a “page skeleton” to save tokens, and clicking by visible text for robustness.

• coding-agent/
  A coding helper agent that works against a local workspace with simple tools.

• data-analyst-agent/
  Experiments with analysis-style workflows (explore → summarize → explain).

• email-agent/
  Agent patterns for drafting, rewriting, and iterating on emails.

• langgraph-learnings/
  Notes + experiments while learning LangGraph patterns.

• mcp-project-manager/
  Experiments around MCP-style tooling and project-management workflows.

• multi-agent-pipeline/
  Experiments chaining multiple “roles” together (planner/researcher/writer, etc.).

• rag-cli/
  RAG experiments packaged as a small CLI-style project.

• research-agent/
  Research workflow patterns: gather sources → filter → synthesize.

• web-search-agent/
  A simple agent that uses web search as a tool to answer questions.

How to run stuff

Everything is meant to be run per-project. Go into a folder and follow its README.md.

Typical flow (varies a bit by project):

1. cd <project>
2. uv sync
3. copy .env.example to .env and add your API keys
4. run uv run main.py

Notes

• API keys belong in .env files and should not be committed.
• Expect rough edges--this repo is intentionally optimized for learning speed, not perfection.