Completed by Pranav · BBACA Graduate · Modern College of Arts, Science & Commerce, Pune
"The best way to learn AI is to build something with it."
- About This Internship
- Task Summary
- Task 1 — Language Translation Tool
- Task 2 — FAQ Chatbot · ARIA
- Task 4 — Object Detection & Tracking · TITAN
- Setup & Installation
- How to Run
- Project Structure
- What I Learned
- Acknowledgements
- Contact
This repository contains my completed AI projects for the CodeAlpha Artificial Intelligence Internship — a program focused on building real-world AI applications using Python.
Each project covers a distinct domain of AI:
| Domain | Project |
|---|---|
| Natural Language Processing | FAQ Chatbot — ARIA |
| Translation Systems | Language Translation Tool |
| Computer Vision & Deep Learning | Object Detection — TITAN |
| Desktop Application Development | All three projects use Tkinter |
| Task | Project | Status |
|---|---|---|
| Task 1 | Language Translation Tool | ✅ Completed |
| Task 2 | FAQ Chatbot — ARIA | ✅ Completed |
| Task 3 | Music Generation with AI | ⏭ Skipped |
| Task 4 | Object Detection & Tracking — TITAN | ✅ Completed |
Certificate requirement: Minimum 2 tasks completed. ✅
A desktop translation application built entirely in Python — a self-contained mini Google Translate with no paid API key required.
┌─────────────────────────────────────────────────────┐
│ User enters text │
└─────────────────────────┬───────────────────────────┘
│
┌───────────▼────────────┐
│ Select source lang │ ← Auto-detect supported
│ Select target lang │ ← 20+ languages
└───────────┬────────────┘
│
┌───────────▼────────────┐
│ deep-translator │
│ (Google Translate │
│ engine — free) │
└───────────┬────────────┘
│
┌───────────▼────────────┐
│ Translated text │
│ displayed instantly │
└────────────────────────┘
| Feature | Description |
|---|---|
| 🌍 20+ Languages | Hindi, French, Spanish, Japanese, Arabic, German & more |
| 🔍 Auto-detect | Detects source language automatically |
| ⇄ Swap | One-click swap between source and target |
| 📋 Copy | Copy translated output to clipboard |
| 🗑 Clear | Clear both panels in one click |
| 📊 Status bar | Shows translation result live |
| 🌙 Dark UI | Modern dark-themed Tkinter interface |
┌──────────────────────────────────────────────────────┐
│ 🌐 Language Translation Tool CodeAlpha Task 1 │
│ ────────────────────────────────────────────────── │
│ │
│ From [ Auto Detect ▼ ] ⇄ To [ Hindi ▼ ] │
│ │
│ ┌───────────────────┐ ┌───────────────────┐ │
│ │ Enter text... │ │ Translation... │ │
│ │ │ │ │ │
│ │ │ │ │ │
│ └───────────────────┘ └───────────────────┘ │
│ │
│ [ Translate ➜ ] [ 📋 Copy ] [ 🗑 Clear ] │
│ ✅ Translated: English → Hindi │
└──────────────────────────────────────────────────────┘
| Library | Purpose |
|---|---|
| Python | Core language |
| Tkinter | Desktop GUI |
| deep-translator | Google Translate engine (free, no API key) |
ARIA = AI Research & Information Assistant
A desktop chatbot that answers technical questions across 10+ domains using NLP-powered semantic matching — not keyword matching. ARIA understands the meaning of your question, then routes it through a dual-layer intelligence system.
User Question
│
┌───────────────▼────────────────┐
│ LAYER 1 — FAQ Bank │
│ 200 Q&A pairs across 10 domains│
│ TF-IDF Vectorization (bigrams) │
│ + Cosine Similarity matching │
└───────────────┬────────────────┘
│
┌──────────▼──────────┐
│ Score ≥ 0.20 ? │
└────┬────────────────┘
│
┌───────────┴───────────────┐
│ YES │ NO (low confidence)
▼ ▼
┌──────────────┐ ┌──────────────────────┐
│ FAQ Bank │ │ LAYER 2 — Claude │
│ Answer │ │ AI API (dynamic) │
│ [FAQ ✓] │ │ [Claude AI ✨] │
│ (instant) │ │ (any tech question)│
└──────────────┘ └──────────────────────┘
User Input: "how does ML work?"
│
▼
Step 1 — Preprocessing (NLTK)
lowercase → remove punctuation → remove stopwords → stem
"how does ML work?" ──► "ml work"
│
▼
Step 2 — TF-IDF Vectorization (Scikit-learn, bigrams)
Converts text into a weighted numerical vector
TF = how often a word appears in this question
IDF = how rare the word is across all 200 FAQs
│
▼
Step 3 — Cosine Similarity
Compare user vector against all 200 FAQ vectors
Score 1.0 = identical meaning
Score 0.0 = completely unrelated
│
▼
Step 4 — Routing Decision
Score ≥ 0.20 → FAQ answer [FAQ ✓] green badge
Score < 0.20 → Claude API [AI ✨] teal badge
| Domain | Example Questions |
|---|---|
| 🤖 AI & ML | What is deep learning? What is overfitting? |
| 🐍 Python | What is list comprehension? What is a decorator? |
| 🗄️ SQL & Databases | What is a JOIN? What is ACID? What is normalization? |
| ☕ Java | What is the JVM? What is multithreading in Java? |
| 🌐 Web Dev | What is REST API? What is the DOM? What is React? |
| 💡 LLMs & Gen AI | What is ChatGPT? What is RAG? How do LLMs work? |
| ☁️ Cloud & DevOps | What is Docker? What is Kubernetes? What is CI/CD? |
| 🔗 Networking | TCP vs UDP? What is DNS? What is HTTP? |
| 🔒 Cybersecurity | What is SQL injection? What is encryption? |
| 💻 DSA | What is Big O notation? What is a hash table? |
┌──────────────────────────────────────────────┐
│ 🤖 ARIA — FAQ Chatbot CodeAlpha Task 2 │
│ ────────────────────────────────────────── │
│ │
│ ARIA │
│ 👋 Hi! I'm ARIA — AI Research & │
│ Information Assistant. │
│ Ask me anything about AI, ML, Python, │
│ SQL, Java, Web Dev, Cloud & more! │
│ │
│ You │
│ what is machine learning │
│ │
│ ARIA [FAQ ✓] 🤖 AI / ML │
│ Machine Learning is a subset of AI │
│ where systems learn from data without │
│ being explicitly programmed... │
│ │
│ You │
│ explain binary search trees │
│ │
│ ARIA [Claude AI ✨] 💬 General Tech │
│ A Binary Search Tree stores nodes where │
│ left child < parent < right child... │
│ │
│ ────────────────────────────────────────── │
│ [ Type your question here... ] [Send ➜][🗑] │
│ FAQ Bank: 200 Q&As | Claude AI fallback │
└──────────────────────────────────────────────┘
| Library | Purpose |
|---|---|
| Python | Core language |
| NLTK | Tokenization, stopword removal, stemming |
| Scikit-learn | TF-IDF vectorizer + cosine similarity |
| Anthropic API | Claude AI fallback for unknown questions |
| Tkinter | Dark-themed chat interface |
TITAN = Tracking & Identification Technology for Autonomous Networks
A real-time AI object detection and tracking system. Point your webcam at anything — TITAN detects and tracks every object it sees using YOLOv8 and assigns persistent IDs that follow objects across frames.
┌──────────────────────────────────────────────────────┐
│ Webcam / Video File │
└────────────────────────┬─────────────────────────────┘
│ frame-by-frame
▼
┌──────────────────────────────────────────────────────┐
│ YOLOv8 Model │
│ │
│ Divides frame into a grid │
│ Each cell predicts: object class + bounding box │
│ + confidence score │
│ │
│ Detects 80 COCO object classes: │
│ person, phone, laptop, car, bottle, chair... │
└────────────────────────┬─────────────────────────────┘
│ detected boxes + scores
▼
┌──────────────────────────────────────────────────────┐
│ ByteTrack │
│ │
│ Matches detections across frames using IoU │
│ Assigns a persistent tracking ID to each object │
│ Same object keeps same ID even if briefly hidden │
└────────────────────────┬─────────────────────────────┘
│ labeled + tracked frame
▼
┌──────────────────────────────────────────────────────┐
│ OpenCV Display Window │
│ │
│ Bounding box with corner accents per object │
│ Badge: #ID ClassName Confidence% │
│ HUD overlay: FPS · Object count · Model · Controls │
└──────────────────────────────────────────────────────┘
People → person
Vehicles → car, motorcycle, bus, truck, bicycle
Electronics → laptop, phone, TV, keyboard, mouse
Furniture → chair, sofa, bed, dining table
Kitchen → bottle, cup, fork, knife, bowl
Animals → cat, dog, bird, horse
Outdoors → traffic light, stop sign, umbrella
Sports → ball, skateboard, tennis racket
+ many more...
| Model | File | Speed | Accuracy |
|---|---|---|---|
| YOLOv8 Nano | yolov8n.pt | ⚡ Fastest | Good |
| YOLOv8 Small | yolov8s.pt | ⚡⚡ Fast | Better |
| YOLOv8 Medium | yolov8m.pt | ⚡⚡⚡ Moderate | Best |
All model weights (~6–50 MB) auto-download from Ultralytics on first run.
┌──────────────────────────────────────────────────────────┐
│ │
│ ┌─────────────────────────────────┐ │
│ │ #1 person 94% │ ← label badge │
│ ├ · · · · · · · · · · ┤ ← corner accents │
│ │ │ │
│ │ [person detected] │ ← bounding box │
│ │ │ │
│ ├ · · · · · · · · · · ┤ │
│ └─────────────────────────────────┘ │
│ │
│ ┌──────────────────────────┐ ← HUD (top-right) │
│ │ FPS : 29.8 │ │
│ │ Objects: 3 │ │
│ │ Model : YOLOV8N │ │
│ │ [ P ] Pause [ Q ] Quit │ │
│ └──────────────────────────┘ │
│ │
└──────────────────────────────────────────────────────────┘
| Key | Action |
|---|---|
P |
Pause / Resume detection |
Q |
Quit detection window |
┌──────────────────────────────────────────────┐
│ 🎯 TITAN CodeAlpha AI Task 4 │
│ ────────────────────────────────────────── │
│ │
│ 📹 Video Source │
│ ◉ Webcam (index 0) │
│ ○ Video File [Browse…] │
│ │
│ 🧠 YOLO Model │
│ [ YOLOv8 Nano (fastest) ▼ ] │
│ │
│ 🎚️ Confidence Threshold │
│ [━━━━━━━━━━━━━━━━━━━━━━━━━━━━━] 0.45 │
│ │
│ 📊 Live Stats │
│ ┌──────────┬──────────┬───────────┐ │
│ │ FPS │ Objects │ Status │ │
│ │ 29.4 │ 4 │ Running │ │
│ └──────────┴──────────┴───────────┘ │
│ │
│ [▶ Start Detection] [■ Stop] │
│ ────────────────────────────────────────── │
│ [ P ] Pause / Resume [ Q ] Quit │
└──────────────────────────────────────────────┘
| Library | Purpose |
|---|---|
| Python | Core language |
| YOLOv8 (Ultralytics) | Real-time object detection |
| ByteTrack | Persistent object ID tracking across frames |
| OpenCV | Video capture, frame processing, display |
| NumPy | Array math for drawing operations |
| Tkinter | Desktop control panel UI |
- Python 3.9 or higher
pip- Internet connection (model weights + NLTK data auto-download on first run)
pip install deep-translator nltk scikit-learn ultralytics opencv-python numpy anthropic
# Windows set ANTHROPIC_API_KEY=sk-ant-... # Mac / Linux export ANTHROPIC_API_KEY=sk-ant-...
Get a free key at: console.anthropic.com
Without a key, ARIA still works — it uses the FAQ bank for all answers.
Task 1 — Language Translation Tool
cd CodeAlpha_LanguageTranslation
python translator.pyTask 2 — FAQ Chatbot (ARIA)
cd CodeAlpha_FAQChatbot
python chatbot.pyNLTK data downloads automatically on first launch (~2 sec).
Task 4 — Object Detection & Tracking (TITAN)
cd CodeAlpha_ObjectDetection
python detector.pyYOLOv8 weights auto-download on first launch. Select model in UI, click Start Detection.
CodeAlpha-AI-Internship/
│
├── CodeAlpha_LanguageTranslation/
│ ├── translator.py ← run this
│ └── requirements.txt
│
├── CodeAlpha_FAQChatbot/
│ ├── chatbot.py ← run this
│ ├── faqs.py ← 200 Q&A knowledge base
│ └── requirements.txt
│
├── CodeAlpha_ObjectDetection/
│ ├── detector.py ← run this
│ └── requirements.txt
│
└── README.md
Raw text ──► Preprocessing ──► TF-IDF vector ──► Cosine score ──► Answer
"how ML?" lowercase 200-dim ≥ 0.20? FAQ
stopwords bigrams < 0.20? Claude AI
stemming
- Text preprocessing pipeline: tokenization → stopwords → stemming
- Why TF-IDF captures importance, not just frequency
- How cosine similarity measures semantic closeness regardless of text length
- When to use retrieval (FAQ) vs generation (Claude AI) — and how to combine both
Frame ──► YOLO grid ──► Box predictions ──► ByteTrack ──► Display
image NxN cells class + conf IoU match annotated
each predicts bounding box persistent ID frame
- How YOLO divides a frame into a grid and predicts multiple boxes per cell
- What confidence threshold does and how tuning it affects detection sensitivity
- How ByteTrack uses Intersection over Union (IoU) to match detections across frames
- Trade-offs between Nano (speed) vs Medium (accuracy) model sizes
- Threading — keeping the Tkinter UI responsive while OpenCV runs on a background thread
- OOP — separating
DetectionSessionfromControlPanelwith clean callbacks - Threading with
daemon=Trueandroot.after()for safe UI updates try/except ImportErrorpattern for graceful dependency handling- Structuring a multi-file project for clean GitHub submission
- CodeAlpha — for the internship opportunity and well-structured task list
- Ultralytics — for YOLOv8, the cleanest object detection API available
- Anthropic — for Claude AI powering ARIA's dynamic fallback
- NLTK Team — for the NLP toolkit behind ARIA's preprocessing
- Scikit-learn — for TF-IDF and cosine similarity
Pranav BBACA Graduate Modern College of Arts, Science and Commerce, Pune-16
Built with 🧠 + ☕ during the CodeAlpha AI Internship