Configure subfolder context root and automated actions pipeline

.github/workflows/deploy_to_huggingface.yml

@@ -0,0 +1,28 @@
+name: Sync to Hugging Face Space
+
+on:
+  push:
+    branches: [main]
+  workflow_dispatch:
+
+jobs:
+  deploy-to-space:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
+
+      - name: Configure git identity
+        run: |
+          git config user.email "github-actions[bot]@users.noreply.github.com"
+          git config user.name "github-actions[bot]"
+
+      - name: Push to Hugging Face Space
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+        run: |
+          git push --force https://ajmel:$HF_TOKEN@huggingface.co/spaces/ajmel/multi-content-engine main

.gitignore

@@ -1,9 +1,10 @@
-venv
+venv/
 .env
-__pycache__/
-*.pyc
+**/__pycache__/
+**/*.pyc
+**/*.pyo
 instance/
 *.log
-chroma_db
+chroma_db/
+.pytest_cache/
 rag-system/test/loade.py
-

README.md

@@ -1,276 +1,103 @@
-# 🤖 AI Engineering Portfolio
-
-A structured, end-to-end AI engineering roadmap covering three production-focused projects — from document intelligence and multimodal processing to AI safety and red-teaming. Each project is built with real architecture, not tutorial code.
-
 ---
-
-## 📂 Repository Structure
-
-```
-├── rag-system/          # Phase 1 — Chat with PDF (RAG Pipeline)
-├── multimodal-engine/   # Phase 2 — AI Reel Generator (Multimodal)
-├── ai-saftey-audit/     # Phase 3 — Safety & Bias Evaluation Toolkit
-├── requirements.txt     # Shared top-level dependencies
-├── .env
-├── .gitignore
-└── README.md
-
-```
-
+title: multi-content-engine
+emoji: 🎬
+colorFrom: blue
+colorTo: indigo
+sdk: docker
+app_port: 7860
+app_dir: multimodal-engine
 ---
 
-## 🗺️ Project Overview
-
-### Phase 1 — RAG System ✅ Complete
+<div align="center">
 
-**`/rag-system`**
+# 🤖 AI Engineering Portfolio
 
-A production-grade Retrieval-Augmented Generation pipeline. Upload any PDF and have a grounded, citation-backed conversation with its contents. Features a multi-LLM fallback chain (Ollama → HuggingFace → OpenAI → Google Gemini), persistent ChromaDB vector storage, sentence-aware chunking, and a full Streamlit chat UI.
+**Three production-grade AI systems — built end-to-end, deployed, and tested.**
 
-🔗 **Live Demo:** [ragsystem-chatpadf.streamlit.app](https://ragsystem-chatpadf.streamlit.app/)
+[![Phase 1 — Live](https://img.shields.io/badge/Phase_1-Live_on_Streamlit-FF4B4B?logo=streamlit&logoColor=white)](https://ragsystem-chatpadf.streamlit.app/)
+[![Phase 2 — Live](https://img.shields.io/badge/Phase_2-Live_on_HuggingFace-FFD21E?logo=huggingface&logoColor=black)](https://huggingface.co/spaces)
+[![Python](https://img.shields.io/badge/Python-3.12-3776AB?logo=python&logoColor=white)](https://python.org)
+[![Tests](https://img.shields.io/badge/Tests-72_passing-brightgreen?logo=pytest)](multimodal-engine/test/)
 
-**Status:** Fully functional — live deployed on Streamlit Cloud with UI, CLI runner, and diagnostic test suite.
+</div>
 
 ---
 
-### Phase 2 — Multimodal Engine 🔧 In Progress
-
-**`/multimodal-engine`**
-An AI engine designed to process, analyze, and generate content across multiple data types — text, images, and audio. The core objective is building automated pipelines that link transcription, summarization, and video processing into a single workflow.
+## 📂 Projects
 
-**Planned Deliverable:** AI Reel Generator — takes long-form video/audio, transcribes it, identifies key moments, and outputs a short-form highlight reel.
-
-**Status:** Architecture planned. Implementation in progress.
+| # | Project | What It Does | Stack | Status |
+|---|---|---|---|---|
+| 1 | [**RAG System**](./rag-system/) | Chat with any PDF — grounded answers with page citations | Gemini · ChromaDB · LangChain · Streamlit | ✅ [Live](https://ragsystem-chatpadf.streamlit.app/) |
+| 2 | [**Multimodal Engine**](./multimodal-engine/) | Convert long videos into vertical reels + blog posts | Gemini 2.5 · FFmpeg · Pydantic · Streamlit | ✅ [Live](https://huggingface.co/spaces) |
+| 3 | [**AI Safety Audit**](./ai-saftey-audit/) | LLM red-teaming, bias scanning, prompt injection defense | Custom framework · ragas | 🔧 In progress |
 
 ---
 
-### Phase 3 — AI Safety Audit 🔧 In Progress
+## 🗂️ Structure
 
-**`/ai-saftey-audit`**
-
-An alignment and auditing framework for stress-testing LLMs against adversarial inputs. Covers prompt injection defense, bias and fairness auditing, toxicity guardrails, and jailbreak red-teaming.
-
-**Planned Deliverable:** AI Safety Audit Report + automated testing framework.
-
-**Status:** Architecture planned. Implementation in progress.
-
----
-
-## 🏗️ Architecture Overview
-
-```text
-┌─────────────────────────────────────────────────────────────────┐
-│                     AI ENGINEERING PORTFOLIO                    │
-├─────────────────────────────────────────────────────────────────┤
-│                                                                 │
-│  Phase 1: RAG System                                            │
-│  ┌──────────┐   ┌──────────┐   ┌──────────┐   ┌────────────┐  │
-│  │  loader  │──►│ chunker  │──►│embedding │──►│ ChromaDB   │  │
-│  └──────────┘   └──────────┘   └──────────┘   └─────┬──────┘  │
-│                                                      │         │
-│  ┌──────────┐   ┌─────────��┐   ┌──────────────────┐ │         │
-│  │Streamlit │◄──│qa_pipeln │◄──│    retriever     │◄┘         │
-│  │   UI     │   │(LLM chain│   │ (distance filter)│           │
-│  └──────────┘   └──────────┘   └──────────────────┘           │
-│                                                                 │
-├─────────────────────────────────────────────────────────────────┤
-│                                                                 │
-│  Phase 2: Multimodal Engine (Planned)                           │
-│  ┌──────────┐   ┌──────────┐   ┌──────────┐   ┌────────────┐  │
-│  │  Video   │──►│ Whisper  │──►│Summarize │──►│Reel Output │  │
-│  │  Input   │   │(Transcr.)│   │  (LLM)   │   │ Generator  │  │
-│  └──────────┘   └──────────┘   └──────────┘   └────────────┘  │
-│                                                                 │
-├─────────────────────────────────────────────────────────────────┤
-│                                                                 │
-│  Phase 3: AI Safety Audit (Planned)                             │
-│  ┌──────────┐   ┌──────────┐   ┌──────────┐   ┌────────────┐  │
-│  │ Prompt   │──►│ Injection│──►│  Bias    │──►│   Audit    │  │
-│  │ Red-Team │   │ Defense  │   │ Scanner  │   │   Report   │  │
-│  └──────────┘   └──────────┘   └──────────┘   └────────────┘  │
-│                                                                 │
-└─────────────────────────────────────────────────────────────────┘
 ```
-
----
-
-## ✅ Features by Project
-
-### RAG System
-
-- PDF text extraction with regex-based cleaning (pypdf)
-- NLTK sentence-aware chunking (no mid-sentence cuts)
-- Semantic embeddings via `all-MiniLM-L6-v2` (HuggingFace)
-- Persistent ChromaDB vector store
-- Distance-filtered retrieval (cosine similarity threshold)
-- Multi-LLM fallback: Ollama → HuggingFace → OpenAI → Google Gemini
-- Strictly grounded prompt — LLM cannot answer outside retrieved context
-- Page-level citations with text snippets and distance scores
-- Full Streamlit chat UI with session state management
-- Chunk quality diagnostic tooling
-
-### Multimodal Engine _(planned)_
-
-- Audio transcription via OpenAI Whisper
-- LLM-powered content summarization
-- Automated video timeline slicing
-- Cross-modal search (text query → image/video results)
-- Unified API for vision and language models
-
-### AI Safety Audit _(planned)_
-
-- Prompt injection detection and filtering
-- Automated bias and fairness test suites
-- Toxicity evaluation layer
-- Jailbreak red-teaming framework
-- Structured audit report generation
-
----
-
-## 🛠️ Tech Stack
-
-| Category       | Technologies                                             |
-| -------------- | -------------------------------------------------------- |
-| Language       | Python 3.10+                                             |
-| UI             | Streamlit                                                |
-| LLM — Local    | Ollama (llama3)                                          |
-| LLM — Cloud    | OpenAI GPT-4o-mini, Google Gemini 2.5 Flash              |
-| Embeddings     | HuggingFace `sentence-transformers` (`all-MiniLM-L6-v2`) |
-| Vector DB      | ChromaDB                                                 |
-| PDF Parsing    | pypdf                                                    |
-| Text Splitting | LangChain, NLTK                                          |
-| Audio/Video    | OpenAI Whisper _(Phase 2)_                               |
-| Safety/Eval    | Custom framework + `ragas` _(Phase 3)_                   |
-| Environment    | python-dotenv                                            |
-| Security       | cryptography, pyjwt                                      |
-
----
-
-## 📦 Installation
-
-**1. Clone the repository**
-
-```bash
-git clone https://github.com/ajme-abes/RAG-Multimodal-SafeAI.git
-cd your-repo
+├── rag-system/           # Phase 1 — RAG Pipeline
+├── multimodal-engine/    # Phase 2 — Multimodal Content Engine
+├── ai-saftey-audit/      # Phase 3 — AI Safety Toolkit
+├── .env.example
+├── .gitignore
+└── README.md
 ```
 
-**2. Create and activate a virtual environment**
-
-```bash
-python -m venv venv
-
-# Windows
-venv\Scripts\activate
-
-# macOS / Linux
-source venv/bin/activate
-```
+> Each project has its own detailed README with architecture, setup, and usage guides.
 
-**3. Install shared dependencies**
+---
 
-```bash
-pip install -r requirements.txt
-```
+## Phase 1 — RAG System
 
-**4. Set up environment variables**
+**[→ Full README](./rag-system/README.md) · [→ Live Demo](https://ragsystem-chatpadf.streamlit.app/)**
 
-Create a `.env` file in the root directory:
+Upload any PDF and have a grounded, citation-backed conversation with it. Multi-LLM fallback chain (Ollama → HuggingFace → OpenAI → Gemini), persistent ChromaDB vector store, sentence-aware chunking, and streaming responses.
 
-```env
-OPENAI_API_KEY=your_openai_key_here
-GOOGLE_API_KEY=your_google_key_here
-HF_API_KEY=your_huggingface_key_here
-```
+| | |  | |
+|---|---|---|---|
+| ![Dashboard](rag-system/assets/image.png) | ![Upload](rag-system/assets/image_pdfup.png) | ![Chat](rag-system/assets/image_chat.png) | ![Score](rag-system/assets/image_scor.png) |
 
 ---
 
-## 🚀 Usage
-
-### Run the RAG System
-
-```bash
-cd rag-system/app
-streamlit run app.py
-```
-
-See [`rag-system/README.md`](./rag-system/README.md) for full setup and usage details.
-
-### Run Diagnostic Tests
+## Phase 2 — Multimodal Content Engine
 
-```bash
-# Chunk quality audit
-python rag-system/test/inspect_chunk.py
+**[→ Full README](./multimodal-engine/README.md) · [→ Live Demo](https://huggingface.co/spaces)**
 
-# Semantic search test
-python rag-system/test/test_search.py
+Upload an MP4 → get a 9:16 vertical reel and a CMS-ready blog post. Parallel audio/visual processing tracks, two-stage AI clip verification, and dual FFmpeg render modes.
 
-# Vector DB connection test
-python rag-system/test/testdb_load.py
-```
+| | | |
+|---|---|---|
+| ![Dashboard](multimodal-engine/assets/dashboard.png) | ![Blog](multimodal-engine/assets/mdpage.png) | ![Reels](multimodal-engine/assets/reelpage.png) |
 
 ---
 
-## 🖼️ Screenshots
+## Phase 3 — AI Safety Audit
 
-**RAG System** — Live at [ragsystem-chatpadf.streamlit.app](https://ragsystem-chatpadf.streamlit.app/)
-
-| Dashboard | Document Ingestion | Chat + Citations |
-|---|---|---|
-| ![Dashboard](rag-system/assets/image.png) | ![Ingestion](rag-system/assets/image_pdfup.png) | ![Chat](rag-system/assets/image_chat.png) |
+**[→ Full README](./ai-saftey-audit/README.md)**
 
-> **Multimodal Engine** and **AI Safety Audit** screenshots will be added as each phase ships.
+An alignment and auditing framework for stress-testing LLMs — prompt injection defense, bias and fairness auditing, toxicity guardrails, and jailbreak red-teaming.
 
----
-
-## 📈 Roadmap
-
-### Phase 1 — RAG System
-
-- [x] PDF loader with text cleaning
-- [x] NLTK sentence-aware chunker
-- [x] HuggingFace embedding model
-- [x] ChromaDB persistent vector store
-- [x] Distance-filtered retriever
-- [x] Multi-LLM fallback chain (Ollama / HF / OpenAI / Gemini)
-- [x] Streamlit chat UI with citations
-- [x] Conversation memory (chat history in prompt)
-- [x] Streaming LLM responses
-- [x] Multi-document support
-- [x] Confidence gate (block hallucination on off-topic queries)
-- [ ] Cross-encoder reranking
-- [ ] FastAPI backend
-- [ ] Docker deployment
-
-### Phase 2 — Multimodal Engine
-
-- [ ] Whisper audio transcription
-- [ ] LLM summarization pipeline
-- [ ] Video clip extraction
-- [ ] Reel generator output
-- [ ] Cross-modal search API
-
-### Phase 3 — AI Safety Audit
-
-- [ ] Prompt injection test suite
-- [ ] Bias and fairness scanner
-- [ ] Toxicity guardrail layer
-- [ ] Jailbreak red-team framework
-- [ ] Automated audit report generator
+**Status:** Architecture planned. Implementation in progress.
 
 ---
 
-## 🔮 Future Improvements
+## 🛠️ Tech Across All Phases
 
-- **RAG Evaluation** — Integrate `ragas` to score faithfulness, answer relevancy, and context recall across all three phases
-- **Unified API Gateway** — Single FastAPI service exposing all three systems under one interface
-- **Docker Compose** — One-command deployment for the entire portfolio stack
-- **CI/CD Pipeline** — GitHub Actions for automated testing and linting on every push
-- **Async Processing** — Parallelize embedding and inference for production-scale throughput
-- **Web Scraping Ingestion** — Extend the RAG loader to ingest URLs and web pages, not just PDFs
+| Category | Tools |
+|---|---|
+| **Language** | Python 3.12 |
+| **UI** | Streamlit |
+| **LLMs** | Google Gemini 2.5 Flash/Pro · OpenAI GPT-4o-mini · Ollama llama3 |
+| **Embeddings** | HuggingFace `all-MiniLM-L6-v2` |
+| **Vector DB** | ChromaDB |
+| **Audio/Video** | FFmpeg · Google Gemini 2.5 Flash/Pro |
+| **Validation** | Pydantic v2 |
+| **Testing** | pytest — 72 tests, 0 failures |
+| **Deployment** | Streamlit Cloud · Hugging Face Spaces · Docker |
 
 ---
 
 ## 👤 Author
 
-Built as part of a structured AI engineering internship roadmap — progressing from RAG fundamentals through multimodal systems to AI safety and evaluation.
+Built as a structured AI engineering roadmap — progressing from RAG fundamentals through multimodal systems to AI safety and evaluation.

multimodal-engine/.gitignore

@@ -2,9 +2,15 @@ data/*.mp4
 data/*.mp3
 data/*.wav
 data/*.avi
-data/extracted_frames
-data/extracted_clips
-data/extracted_real
-output/clips/*.mp4
+data/extracted_frames/
+data/extracted_clips/
+data/extracted_real/
+data/temp_verification_slices/
+data/transcript.json
+output/clips/
+output/*.md
 assets/video.mp4
 multimodal-engine/assets/video.mp4
+__pycache__/
+*.pyc
+.pytest_cache/

multimodal-engine/Dockerfile

@@ -14,15 +14,15 @@ ENV HOME=/home/user \
 
 WORKDIR $HOME/app
 
-# Copy local dependency maps into virtual container layers
+# Since app_dir enters the subfolder, local paths are relative to multimodal-engine/
 COPY --chown=user requirements.txt .
 RUN pip install --no-cache-dir -r requirements.txt
 
-# Copy the rest of the application codebase
+# Copy the specific app folder contents safely
 COPY --chown=user . .
 
-# Hugging Face Spaces strictly requires port 7860
+# Hugging Face Spaces strictly routes traffic over port 7860
 EXPOSE 7860
 
-# Run Streamlit on Hugging Face's required port configuration layout
-ENTRYPOINT ["streamlit", "run", "multimodal-engine/app/app.py", "--server.port=7860", "--server.address=0.0.0.0"]
+# Launch Streamlit on the specified port
+ENTRYPOINT ["streamlit", "run", "app/app.py", "--server.port=7860", "--server.address=0.0.0.0"]