• Built a planner → architect → coder LangGraph pipeline that turns plain-language specs into runnable project scaffolds with controlled recursion.
• Integrated Groq’s openai/gpt-oss-120b through LangChain with Pydantic outputs to keep task plans deterministic.
• Hardened the tool layer for sandboxed file and command access, enabling safe, automated generation of multi-file web app prototypes.

Github Link

• Built a knowledge-graph backend combining FastAPI with Ollama/OpenAI pipelines for automatic entity and relationship extraction from unstructured documents.
• Engineered secure, production-grade ingestion (MIME validation, path traversal defense, suspicious-content scanning) with version-controlled graph history and REST APIs for upload, retrieval, and lifecycle management.
• Optimized persistence via SQLAlchemy models, Alembic migrations, and batched operations for low-latency queries plus detailed stats/monitoring endpoints.

Github Link

• Developed a deep learning-based IDS using the NSL-KDD dataset, optimizing feature selection with RFE and a Decision Tree classifier.
• Achieved 95% accuracy and 0.94 F1-score with a CNN-LSTM model; work accepted in Elsevier Computers & Security (Q1).

Github Link

• Leveraged a pretrained ResNet50 convolutional backbone with Global Max Pooling to encode images into dense feature embeddings.
• Applied L2 normalization and scikit-learn’s brute-force kNN (Euclidean) search to rank visually similar catalog items.
• Created a Streamlit interface that uploads customer images, infers embeddings on the fly, and displays the top recommendations in real time.

Github Link