Vivek Tippireddy

Description: A hybrid multi-agent research architecture that uses a local Qwen 3 0.6B model (via Ollama) for planning, search orchestration, and research synthesis, reserving Gemini exclusively for final report generation.

Key Features:

• Multi-stage LangGraph workflow: task decomposition, tool/search orchestration, research synthesis, and final answer generation
• Structured-output handling with JSON parsing and fallback mechanisms to improve reliability of smaller local models
• Different agent tasks assigned to different model capabilities instead of a frontier model throughout
• Reduced API requests from 70 to 18 (~74%) and input tokens from ~150K to ~29.9K (~80%)
• Increased observed task success rate from 50% to 83.33%

Technologies: Python, LangGraph, Ollama, Qwen 3 0.6B, Gemini, Web Search

Description: An agentic RAG architecture where a reasoning agent dynamically coordinates multiple information sources, fusing evidence before verification rather than relying on a single retrieval step.

Key Features:

• Unified research workflow integrating vector retrieval, knowledge graphs, SQL/API sources, and web search
• Evidence-fusion layer combining information from different sources before passing it downstream
• Dedicated Verifier/Critic stage for reliability checks
• Distinct stages for reasoning, retrieval, evidence aggregation, verification, and final response generation
• Explores how multi-source evidence and explicit verification improve reliability over single-step RAG

Technologies: Python, RAG, Vector DB, Knowledge Graph, SQL/API, Web Search

Description: A personalized memory architecture that moves beyond retrieving similar historical messages toward generating a compact representation of a user's current goals, preferences, important facts, and recent changes.

Key Features:

• Identifies limitations of retrieval-based memory: outdated information, contradictions, changing goals, and irrelevant historical context
• Memory-processing layer that interprets long-term interaction history and generates context optimized for downstream LLM reasoning
• Explores improvements in personalization, contradiction handling, context-window efficiency, and long-term conversational coherence
• Research question: whether an understanding-oriented memory model provides more useful context than conventional retrieval-based memory

Status: Architecture / Research Prototype

Technologies: AI Systems Architecture, LLMs, Retrieval, Memory