Skip to content

About

πŸ€– AI Engineer Portfolio β€” 8 production-ready projects covering the full AI engineering stack: LLM training, RAG, agents, reasoning, image generation, MCP

Resources

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

AI Engineer Portfolio

Production-grade, cloud-native AI applications covering the full spectrum of modern AI engineering β€” from LLM fundamentals to multi-agent systems with MCP & A2A protocols.

CI Pipeline Python 3.11+ License: MIT

Architecture

ai-engineer-portfolio/
β”œβ”€β”€ libs/common/              # Shared utilities, config, models
β”œβ”€β”€ projects/
β”‚   β”œβ”€β”€ 01-llm-playground/          # LLM fundamentals & text generation
β”‚   β”œβ”€β”€ 02-customer-support-chatbot/ # Fine-tuning, PEFT/LoRA, prompts
β”‚   β”œβ”€β”€ 03-ask-the-web-agent/       # Perplexity-like search agent
β”‚   β”œβ”€β”€ 04-deep-research/           # Reasoning & inference-time scaling
β”‚   β”œβ”€β”€ 05-image-generation/        # Diffusion models & T2I service
β”‚   β”œβ”€β”€ 06-capstone-multiagent/     # Multi-agent orchestration platform
β”‚   β”œβ”€β”€ 07-agent-rag/              # Advanced RAG with agent patterns
β”‚   └── 08-mcp-a2a/               # MCP servers & A2A protocol
β”œβ”€β”€ k8s/                      # Kubernetes manifests
β”œβ”€β”€ .github/workflows/        # CI/CD pipelines
└── docker-compose.yml        # Local development environment

Projects

Project 1: LLM Playground

Topics: LLM Foundations, Pre-Training, Data Collection, Tokenization (BPE), Transformer Architecture, Text Generation

Interactive playground for exploring LLM fundamentals. Compare tokenizers (tiktoken, HuggingFace), experiment with generation strategies (greedy, beam search, top-k, top-p), and explore transformer architectures.

Tech: FastAPI, tiktoken, HuggingFace Tokenizers, Anthropic SDK, OpenAI SDK, SSE streaming

Project 2: Customer Support Chatbot

Topics: Adaptation Techniques, Fine-tuning, PEFT, LoRA, Adapters, Prompt Engineering

Production chatbot with advanced prompt engineering, intent classification, conversation management, and a complete LoRA fine-tuning pipeline. Features versioned prompt templates and conversation state machine.

Tech: FastAPI, LangChain, Redis, PostgreSQL, PEFT/LoRA, Jinja2 templates

Project 3: Ask-the-Web Agent (Perplexity Clone)

Topics: Agents Overview, Agency Levels, Workflow, Prompt Chaining, Routing

Perplexity-like search agent built with LangGraph. Features query routing, web search, content synthesis with citations, and fact checking. Demonstrates agentic patterns: prompt chaining, conditional routing, parallel execution.

Tech: FastAPI, LangGraph, Tavily Search, Anthropic SDK, SSE streaming

Project 4: Deep Research Capability

Topics: Reasoning LLMs, OpenAI o-series, DeepSeek-R1, Inference-time Techniques, CoT Prompting

Deep research engine with Chain-of-Thought and Tree-of-Thought reasoning. Decomposes complex questions, executes multi-step research with iterative deepening, and generates structured reports with confidence scoring.

Tech: FastAPI, LangGraph, Claude Extended Thinking, Inference-time Scaling

Project 5: Image Generation Service

Topics: VAE, GANs, Auto-regressive Models, Diffusion Models, Text-to-Image (T2I)

Cloud-native image generation service with multi-provider support (DALL-E, Stable Diffusion, FLUX). Features prompt enhancement, batch generation, image-to-image, and educational modules explaining generative architectures.

Tech: FastAPI, OpenAI DALL-E, Replicate, Pillow, S3 Storage

Project 6: Capstone Multi-Agent Platform

Topics: Multi-Agent Orchestration, Supervisor Pattern, Shared Memory, Task Decomposition

Capstone project combining all techniques. Multi-agent platform with specialized agents (researcher, coder, analyst, writer) orchestrated by a supervisor via LangGraph. Features shared memory, task decomposition, and parallel execution.

Tech: FastAPI, LangGraph, PostgreSQL, Redis, Multi-Agent Orchestration

Module 7: Agent & RAG System

Topics: Query Decomposition, Context Engineering, Memory, Meta-agent vs Planner vs Orchestrator, Single/Multi/Hierarchical RAG

Comprehensive RAG system implementing three architectures: single-agent, multi-agent, and hierarchical (A-RAG). Features document ingestion, semantic chunking, hybrid retrieval, reranking, and context engineering.

Tech: FastAPI, LangGraph, Qdrant, Cohere, LangChain Text Splitters

Module 8: MCP & A2A Integration

Topics: Model Context Protocol, MCP Servers/Clients, Tool Discovery, A2A Protocol, AgentCards

Full MCP server and client implementation with tool/resource/prompt support. A2A protocol with AgentCards for agent discovery and inter-agent communication. Includes security module for prompt injection defense.

Tech: FastAPI, MCP SDK, A2A Protocol, AgentCards, SSE

Tech Stack

Category Technologies
Language Python 3.11+
API Framework FastAPI, SSE, WebSocket
LLM Providers Anthropic (Claude 4.5/4.6), OpenAI (GPT-4o), Open-source (Llama, DeepSeek)
Agent Frameworks LangGraph, LangChain
Vector Database Qdrant
Databases PostgreSQL, Redis
Containerization Docker, Docker Compose
Orchestration Kubernetes, Kustomize
CI/CD GitHub Actions
Package Management uv (workspace monorepo)
Code Quality Ruff, mypy, pytest
Protocols MCP (Model Context Protocol), A2A (Agent-to-Agent)

Quick Start

Prerequisites

  • Python 3.11+
  • Docker & Docker Compose
  • uv (Python package manager)

1. Clone & Setup

git clone https://github.com/samuelvinay91/ai-engineer-portfolio.git
cd ai-engineer-portfolio
cp .env.example .env
# Edit .env with your API keys

2. Run with Docker Compose (recommended)

docker compose up -d

Services will be available at:

Service URL
LLM Playground http://localhost:8001
Customer Support Chatbot http://localhost:8002
Ask-the-Web Agent http://localhost:8003
Deep Research http://localhost:8004
Image Generation http://localhost:8005
Capstone Multi-Agent http://localhost:8006
Agent RAG http://localhost:8007
MCP & A2A http://localhost:8008

3. Run Locally (individual project)

# Install dependencies
uv sync --all-packages

# Run a specific project
uv run uvicorn projects.01-llm-playground.src.llm_playground.main:app --port 8001

4. Run Tests

# All tests
uv run pytest

# Specific project
uv run pytest projects/01-llm-playground/tests/ -v

5. Deploy to Kubernetes

kubectl apply -k k8s/base/
# Then apply individual project manifests
kubectl apply -f projects/01-llm-playground/k8s/

Models Used

Use Case Recommended Model Alternative
General Chat Claude Sonnet 4.5 GPT-4o
Complex Reasoning Claude Opus 4.6 OpenAI o3
Code Generation Claude Sonnet 4.5 DeepSeek R1
Embeddings Cohere embed-v4 OpenAI text-embedding-3-large
Image Generation DALL-E 3 / FLUX 1.1 Stable Diffusion 3.5
Reranking Cohere Rerank LLM-based reranking

License

MIT License - see LICENSE for details.

About

πŸ€– AI Engineer Portfolio β€” 8 production-ready projects covering the full AI engineering stack: LLM training, RAG, agents, reasoning, image generation, MCP

Resources

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages