ML Projects Showcase
9 end-to-end projects across ML, DL, NLP, and LLMs. Each includes architecture, dataset, skills, and implementation details.
House Price Prediction
End-to-end regression pipeline on the Ames Housing dataset. Feature engineering, EDA, Ridge/Lasso tuning, and a Streamlit dashboard.
⚙️ Architecture
EDA → Feature Engineering → Train/Test Split → Ridge/Lasso/XGBoost → SHAP Explainability → Streamlit UI
📊 Dataset
Ames Housing Dataset (Kaggle) — 79 features, 1,460 training samples
✅ Key Results
- RMSE < 18,000
- Top 15% Kaggle leaderboard
- SHAP feature importance charts
🛠️ Skills Used
Customer Churn Prediction
Telecom churn classifier with SMOTE class balancing, threshold tuning, SHAP explanations, and a business-ready Streamlit report.
⚙️ Architecture
Data Cleaning → SMOTE Oversampling → Logistic/RF/XGBoost → Threshold Optimization → SHAP Dashboard
📊 Dataset
IBM Telco Customer Churn (Kaggle) — 7,043 customers, 20 features
✅ Key Results
- F1-score 0.89 on minority class
- Business ROI report included
- Threshold tuning demo
🛠️ Skills Used
Movie Recommendation System
Collaborative filtering (SVD + ALS) combined with content-based TF-IDF features. FastAPI serving with Redis caching.
⚙️ Architecture
User-Item Matrix → SVD/ALS → Content TF-IDF → Hybrid Scorer → FastAPI → Redis Cache
📊 Dataset
MovieLens 25M — 25M ratings, 62K movies, 162K users
✅ Key Results
- RMSE 0.87 on test set
- Sub-50ms API response
- Hybrid content + collaborative
🛠️ Skills Used
Image Classifier (ResNet-50)
Transfer learning with frozen ResNet-50 backbone fine-tuned on CIFAR-10. Mixed-precision training, WandB tracking, ONNX export.
⚙️ Architecture
ResNet-50 (frozen) → Custom Head → Fine-tune → Mixed Precision → ONNX Export → FastAPI
📊 Dataset
CIFAR-10 — 60,000 images, 10 classes (32×32 px)
✅ Key Results
- 94.3% test accuracy
- ONNX optimized for 10ms inference
- WandB experiment tracking
🛠️ Skills Used
Object Detection (YOLOv8)
Fine-tuned YOLOv8 on a custom dataset for real-time object detection in video streams. Deployed with FastAPI + WebSocket streaming.
⚙️ Architecture
YOLOv8 Nano → Custom Dataset → Fine-tune → TensorRT → FastAPI WebSocket → Browser
📊 Dataset
COCO subset + custom annotated dataset (Roboflow) — 5,000 images
✅ Key Results
- mAP@0.5 = 0.82
- Real-time 30FPS inference
- WebSocket live stream
🛠️ Skills Used
Sentiment Analysis API
Fine-tuned DistilBERT on SST-2 with quantization (INT8) for 3× faster inference. Full REST API with batch endpoint.
⚙️ Architecture
DistilBERT → Fine-tune SST-2 → INT8 Quantization → FastAPI → Docker → Render
📊 Dataset
Stanford SST-2 — 67,349 movie review sentences, binary sentiment
✅ Key Results
- 92.1% accuracy
- 3× faster with INT8 quantization
- Batch inference endpoint
🛠️ Skills Used
RAG Document Chatbot
PDF Q&A using LangChain + FAISS + Groq Llama 3. Semantic chunking, hybrid search (BM25 + dense), and streaming responses.
⚙️ Architecture
PDF → Semantic Chunking → FAISS + BM25 → Reranker → Groq Llama 3 → Streaming API → Next.js
📊 Dataset
User-uploaded PDFs — research papers, textbooks, documentation
✅ Key Results
- Hybrid BM25 + dense retrieval
- Real-time streaming
- Reranking with cross-encoder
🛠️ Skills Used
PDF Q&A with Citations
Upload any PDF and get cited answers. Page-level citation tracking, chunk-level provenance, and confidence scoring.
⚙️ Architecture
PDF → pdfplumber → Page Chunks → Embeddings → FAISS → LLM + Citation Injection → React UI
📊 Dataset
User-provided PDFs (research papers, legal docs, textbooks)
✅ Key Results
- Page-level citations
- Confidence scoring
- Multi-PDF cross-search
🛠️ Skills Used
AI Research Assistant
Multi-agent system for literature review. Agents search ArXiv, summarize papers, extract key ideas, and generate a structured report.
⚙️ Architecture
Query → ArXiv Search Agent → Paper Summarizer Agent → Synthesis Agent → Report Generator → Markdown Export
📊 Dataset
ArXiv API (live) — 2M+ papers across CS/ML/AI
✅ Key Results
- Multi-agent LangGraph workflow
- Live ArXiv search
- Structured PDF report output
🛠️ Skills Used
Ready to build your own ML projects?
Complete the curriculum, earn XP, and apply every concept in a real project.
Start Learning → Earn XP