This repository is my primary workspace for building GenAI and agentic systems.
It contains:
- learning experiments
- foundational implementations
- agent architectures
- tool integrations
- real projects
This is not a single project repo. It is a long-term, evolving engineering workspace focused on how modern AI agents are actually built.
This repo is designed to cover everything required to build real AI agents, including:
- Large Language Model (LLM) usage
- Tool / function calling
- Agent control loops
- Web-augmented reasoning
- Frameworks (LangChain, LangGraph, etc.)
- Retrieval-Augmented Generation (RAG)
- Vector databases
- Observability & evaluation
- Production-oriented patterns
Some projects live inside this repo.
Some mature projects are moved to separate GitHub repositories.
This repo remains the core lab and knowledge base.
Most GenAI repositories:
- jump straight into frameworks
- hide logic behind abstractions
- optimize for demos
This repo follows the opposite approach:
- start from first principles
- build things manually first
- understand control flow
- add frameworks only when their value is clear
The goal is engineering depth, not surface-level familiarity.
genai-agentic-systems/
├── 1-inVoke_LLM/
├── 2-toolCalling/
├── 3-llm_webSearchTool_integreaton/
├── 4-chatbot_project/
├── (future: RAG, vectorDBs, frameworks, evals)
Each directory represents a clear capability or learning stage, not random experiments.
Focus: Direct interaction with LLMs.
Covers:
- invoking LLMs via SDKs
- system vs user prompting
- structured JSON outputs
- deterministic responses
- using LLMs as evaluators (graders, classifiers)
This is the foundation layer of all GenAI systems.
Focus: Understanding how tool/function calling works internally.
Covers:
- why LLMs cannot access external systems
- how models request tools
- how developers execute tools
- the LLM → tool → LLM feedback loop
This folder exists to build correct mental models.
Focus: A real, working AI agent.
Covers:
- interactive CLI agent
- persistent conversation state
- real web search integration (Tavily)
- multi-step reasoning
- safety mechanisms (tool iteration limits)
This is a minimal but real agent system, not a demo.
Alongside learning modules, this repository also contains end-to-end GenAI applications that demonstrate how agentic systems are applied in practice.
Some projects remain here for architectural clarity; others may later be split into standalone repositories.
Location: chatbot_project/
Type: Full-stack Agentic Application
Maya AI is a lightweight, full-stack agentic chatbot that demonstrates how an LLM can reason, decide, and act by invoking external tools for real-time information.
Unlike traditional chatbots, Maya follows an explicit ReAct (Reason + Act) loop, where the LLM decides when web access is required and the system executes tools on its behalf.
- Agentic reasoning with LLMs
- Tool calling for real-time web data (Tavily)
- LLM → Tool → LLM feedback loop
- Separation of reasoning and execution
- Grounded responses using live data
- Minimal frontend + Express backend integration
- Frontend: HTML, Vanilla JavaScript, Tailwind CSS
- Backend: Node.js, Express.js
- LLM: Groq SDK (
llama-3.3-70b-versatile) - Tools: Tavily Web Search API
This project serves as a practical proof-of-work for building real agentic systems without heavy frameworks.
Refer to chatbot_project/README.md for implementation details.
This repo will expand to include:
- Vector databases (Pinecone, FAISS, etc.)
- Retrieval-Augmented Generation (RAG)
- Agent frameworks (LangChain, LangGraph)
- Observability & tracing (Langfuse)
- Evaluation pipelines
- Memory & context management
- Multi-agent orchestration
Each addition will follow the same rule:
understand → implement → abstract
- Experiments & learning → stay inside this repo
- Polished applications → moved to separate repos
- This repo always remains the source of truth for concepts and architecture
- Recruiters evaluating GenAI engineering capability
- Clients looking for agent system builders
- Engineers interested in how agents actually work
- Anyone who cares about clarity over hype
This repository represents a serious, long-term investment in GenAI and agentic systems.
It is:
- structured
- intentional
- engineering-first
- built to scale with new tools and ideas
This is not about “learning GenAI”.
This is about building agent systems properly.