A collection of bite-sized projects that together showcase a full AI-Engineer skill-set — from RAG and LLM gateways to SQL ETL and Vertex AI notebooks.
| Module | Folder | What it proves |
|---|---|---|
| M1 – RAG FAQ Bot | modules/m1-rag-faq |
Embeddings, retrieval, LangChain basics |
| M2 – LLM API Gateway | modules/m2-api-gateway |
FastAPI ↔ OpenAI / Vertex, production patterns |
| M3 – Auto-Reporter Agent | modules/m3-auto-reporter |
LangChain Agent, BigQuery, Slack integration |
| M4 – n8n Flow | modules/m4-n8n |
Low-code orchestration (Webhook → HTTP → Slack) |
| M5 – SQL + ETL | modules/m5-sql-etl |
Window functions, scheduling, data-engineering chops |
| M6 – Vertex Notebook | modules/m6-vertex-notebook |
Gemini 2.5 Flash, GCP ML tooling |
| M7 – Vector-DB Swap | modules/m7-vector-swap |
Pinecone ↔ Chroma hot-swap |
| M8 – CrewAI Agents | modules/m8-crew-agents |
Multi-agent orchestration, memory, feedback loops |
(Each folder contains its own README with 30-sec run instructions & screenshots.)
# 1. Clone & enter
git clone https://github.com/e1washere/company-faq.git
cd company-faq
# 2. Python env
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
# 3. Mandatory keys
export OPENAI_API_KEY="sk-…"
# 4. (Optional) Pinecone creds for live index
export PINECONE_API_KEY="pc-…" # v3 or v2 key
export PINECONE_REGION="us-east-1" # v3, default aws/us-east-1
# OR legacy
env var
export PINECONE_ENV="my-env-id" # v2 environment name
# 5. Build vectors & ask one question
python pinecone_demo.py --rebuild --query "What is Patrianna?"
# 6. Chat interactively
python pinecone_demo.pyNo Pinecone? No FAISS? The script automatically falls back to local Chroma, so it always works out-of-the-box.
graph TD
A[User Input] --> B[M2 API Gateway]
B --> C{Provider Selection}
C -->|OpenAI| D[OpenAI GPT]
C -->|Vertex| E[Vertex AI Gemini]
B --> F[M8 CrewAI Agents]
F --> G[Data Analyst Agent]
F --> H[Report Generator Agent]
F --> I[Root Cause Analyzer]
J[M1 RAG System] --> K[Vector Storage]
K --> L[M7 Vector Swap]
L --> M[Pinecone]
L --> N[Chroma]
L --> O[FAISS]
P[M3 Auto-Reporter] --> Q[BigQuery]
P --> R[Slack Integration]
S[M5 SQL ETL] --> T[Data Warehouse]
U[M6 Vertex Notebook] --> V[Gemini Flash]
W[M4 n8n Workflow] --> X[Low-Code Orchestration]
style F fill:#e1f5fe
style B fill:#fff3e0
style J fill:#f3e5f5
style P fill:#e8f5e8
- Modular Architecture: Each module (M1-M8) is independent and can be deployed separately
- AI-First Design: Every component leverages AI/ML capabilities
- Production-Ready: Comprehensive error handling, logging, and monitoring
- Scalable: Cloud-native design with Docker and Kubernetes support
- Extensible: Easy to add new modules and integrations
sequenceDiagram
participant U as User
participant G as M2 Gateway
participant A as M8 Agents
participant V as Vector Store
participant L as LLM Provider
participant S as Slack
U->>G: Query Request
G->>A: Route to Agent
A->>V: Retrieve Context
V->>A: Return Relevant Data
A->>L: Generate Response
L->>A: AI Response
A->>G: Enhanced Response
G->>U: Final Response
Note over A,S: Async reporting
A->>S: Send Report
graph LR
A[CrewOrchestrator] --> B[Data Analyst Agent]
A --> C[Report Generator Agent]
A --> D[Root Cause Analyzer]
B --> E[Memory System]
C --> E
D --> E
E --> F[Persistent Storage]
E --> G[Context Awareness]
E --> H[Feedback Loops]
style A fill:#ffecb3
style E fill:#c8e6c9
| # | Highlight |
|---|---|
| M1 | LangChain 0.2, ConversationalRetrievalChain, Chroma → Pinecone vector stores |
| M2 | Production-grade REST gateway (FastAPI) with pluggable LLM back-ends + streaming responses |
| M3 | Autonomous agent that queries BigQuery and posts rich Slack reports on schedule |
| M4 | Pure n8n flow file demonstrating low-code webhook → transformation → notification |
| M5 | Advanced SQL (window, CTE) + cron-ready ETL scripts |
| M6 | Vertex AI Workbench notebook running Gemini Flash for rapid prototyping |
| M7 | pinecone_demo.py – swap-able vector-DB layer (Pinecone v3/v2 → FAISS → Chroma) |
| M8 | Multi-agent orchestration with memory, feedback loops, and context awareness |
Deploy the LLM API Gateway to Google Cloud Run in seconds:
# 1. Prerequisites
gcloud auth login
gcloud config set project YOUR_PROJECT_ID
# 2. Deploy in one command
./deploy.sh YOUR_PROJECT_ID us-central1
# 3. Set API keys (after deployment)
gcloud run services update company-faq-api --region=us-central1 \
--set-env-vars="OPENAI_API_KEY=sk-...,VERTEX_PROJECT=YOUR_PROJECT_ID"What gets deployed:
- Production-ready FastAPI server (Module M2) with OpenAI + Vertex AI support
- Auto-scaling from 0 to 10 instances based on traffic
- Built-in monitoring and logging via Cloud Run
- Public HTTPS endpoint with automatic SSL
Example usage:
# Test the deployed API
curl -X POST https://company-faq-api-xxx.a.run.app/chat \
-H "Content-Type: application/json" \
-d '{"messages":[{"role":"user","content":"Hello!"}]}'Requirements:
- Google Cloud SDK installed
- Active GCP project with billing enabled
- Docker (automatically handled by Cloud Build)
| Var | Purpose | Example |
|---|---|---|
OPENAI_API_KEY |
Required for embeddings / chat | sk-… |
PINECONE_API_KEY |
Use managed Pinecone (v3 or v2) | pc-… |
PINECONE_REGION |
v3 region (default us-east-1) |
us-east-1 |
PINECONE_CLOUD |
v3 cloud provider (aws/gcp) |
aws |
PINECONE_ENV / PINECONE_ENVIRONMENT |
v2 environment ID | gcp-starter |
CHROMA_TELEMETRY_ENABLED |
Disable Chroma telemetry | false |
- Python 3.11
- LangChain 0.2 +
langchain_openai - Vector DBs: Pinecone, FAISS, Chroma
- FastAPI, BigQuery, Slack SDK
- n8n (low-code), SQL (BigQuery dialect)
- GCP Vertex AI + Gemini 2.5 Flash
See each folder for 30-sec run instructions & screenshots.
This project directly demonstrates the key requirements from the Patrianna AI Engineer role:
| Requirement | Implementation | Module |
|---|---|---|
| LLM Pipelines (Vertex AI, OpenAI) | Multi-provider API Gateway | M2, M6 |
| RAG & Embedding Solutions | Complete RAG system with vector stores | M1, M7 |
| Agent Orchestration (CrewAI/LangChain) | Multi-agent system with memory | M8 |
| Automated Report Generation | Intelligent reporting with feedback | M3, M8 |
| Root Cause Analysis | Specialized analysis agent | M8 |
| Interactive User Agents | Context-aware conversational agents | M8 |
| API Integrations | Production FastAPI with monitoring | M2 |
| Data Models for AI | Optimized data structures | M1, M3, M5 |
| n8n/Python Workflows | Low-code + Python automation | M4, All |
| SQL for AI-Data Operations | ETL pipelines and analytics | M5, M3 |
| Vector Databases | Pinecone, Chroma, FAISS | M7, M1 |
| Feedback Mechanisms | Memory systems and learning loops | M8, M3 |
| Context-Aware Systems | Persistent memory and insights | M8 |
- Production-Ready: Docker, CI/CD, comprehensive testing
- Scalable Design: Cloud-native with auto-scaling
- Monitoring & Logging: Decision logging and performance tracking
- Error Handling: Graceful degradation and recovery
- Documentation: Comprehensive README and inline docs
- Multi-Agent Orchestration: CrewAI-style agent coordination
- Memory Systems: Persistent agent memory with JSON storage
- Feedback Loops: Continuous improvement mechanisms
- Context Awareness: Historical context influences decisions
- Decision Logging: Audit trail for all AI decisions
- Automated Workflows: End-to-end automation pipelines
- 8 Integrated Modules: Full AI engineering stack
- 34 Unit Tests: Comprehensive test coverage
- Multiple LLM Providers: OpenAI, Vertex AI, Gemini
- 3 Vector Databases: Pinecone, Chroma, FAISS
- Production Deployment: One-command Cloud Run deploy
- Real-time Monitoring: Structured logging and metrics
- Technical Expertise: Advanced AI/ML system design
- Production Skills: Deployment, monitoring, testing
- Architecture Thinking: Scalable, modular design
- Problem Solving: Complex multi-agent orchestration
- Best Practices: Error handling, logging, documentation
- Innovation: Cutting-edge AI agent technologies
This project represents a complete AI engineering portfolio showcasing all skills required for the Patrianna AI Engineer role.