A comprehensive repository documenting the development and optimization of LLM-based applications, RAG pipelines, and Agent Orchestration using LangChain, OpenAI, and Google Gemini.
- Orchestration: LangChain, LangGraph
- Models: OpenAI (GPT-4o/o1), Google Gemini (Pro/Flash)
- Vector Databases: ChromaDB / FAISS
- Language: Python 3.11
The programs in this repository follow a modular architecture designed for scalability and reliability in agentic workflows.
Implementation of core LangChain components:
- Chains: LCEL (LangChain Expression Language) for declarative compositions.
- Memory: Windowed and Summary buffer implementations for stateful conversations.
- Tools: Custom tool-calling wrappers for API and database interactions.
Comparison of model behaviors and integration patterns across top-tier providers:
- OpenAI: Function calling, structured outputs, and fine-tuning scripts.
- Gemini: Multi-modal processing and long-context window utilization.
Advanced retrieval strategies to mitigate hallucinations:
- Ingestion: Recursive character splitting and semantic chunking.
- Retrieval: Hybrid search (Keyword + Vector) and Re-ranking (Cross-Encoders).
- Evaluation: RAGAS metrics for faithfulness and relevancy.
Systematic approaches to prompt construction:
- Few-Shot Learning: Dynamic example selection.
- Chain-of-Thought (CoT): Forcing reasoning paths for complex logic.
- System Messages: Role-based constraint enforcement.
Ensure you have API keys for OpenAI and Google AI Studio.
git clone https://github.com/NIVAS2823/GenAIPractice.git
cd [REPO_NAME]
for every directory there will be separate requirements.txt file
cd #desired directory
pip install -r requirements.txt