Loading projects...
Enterprise RAG Knowledge Assistant
MS Graph Integrated Qdrant Assistant
A sophisticated, enterprise-grade conversational AI platform utilizing Retrieval-Augmented Generation (RAG) to provide instant, citation-backed answers from corporate knowledge bases (OneDrive), featuring autonomous agents and detailed usage analytics.
PythonFastAPILangChainLangGraphMongoDBQdrantUnstructuredMS Graph APIGoogle OAuthDocker
Timeline
2 Weeks
Intensity
Individual
Access
Proprietary
Revision
Mar 2026
The Challenge
Defining the core problem and identified pain points that necessitated this technical intervention.
Enterprises face significant productivity losses due to data silos, where critical organizational knowledge is trapped in unstructured documents (PDFs, DOCX, PPTX) across cloud storage. Employees waste hours manually searching for internal policies, technical documentation, and procedural guides.
Technical Solution
The architectural and implementation strategy developed to resolve the challenge.
01
Advanced RAG Pipeline
Built a robust ingestion engine that syncs with Microsoft OneDrive, utilizing `Unstructured` for high-fidelity document parsing (OCR included) and `Qdrant` for semantic vector search.
02
Agentic Workflow Engine
Implemented a `LangGraph` state machine to orchestrate complex user interactions, enabling the bot to maintain context, clarify ambiguities, and route queries to specialized tools.
03
Dynamic Knowledge Base
Created a live synchronization service (`onedrive_service`) that keeps the vector store up-to-date with file changes, additions, and deletions in real-time.
04
Usage & Cost Analytics
Developed a comprehensive tracking system (`token_usage_router`) to monitor LLM token consumption per user/model, ensuring cost control and visibility.
05
Secure Scalable Architecture
Designed a modular, async architecture with MongoDB for persistence, handling user sessions, chat history, and background tasks efficiently.
Key Contributions
My specific roles, responsibilities, and the technical value I added to the project lifecycle.
Backend Architecture
Architected the complete FastAPI backend using a clean, layered architecture (Router-Service-Core) for maximum maintainability.
RAG Pipeline Development
Developed the core RAG pipeline using LangChain and Qdrant, optimizing embedding strategies for high-accuracy retrieval.
Conversational Agent Graph
Implemented the 'Chatbot Graph' using LangGraph to manage complex conversation states and autonomous tool execution.
OneDrive Sync Service
Built the 'OneDrive Syncer' service, integrating MS Graph API for automated, incremental document synchronization and vectorization.
Token Usage Analytics
Engineered the 'Token Usage' tracking module to provide granular analytics on API costs and model performance.
Document Parsing Pipeline
Integrated 'Unstructured.io' for robust parsing of complex file formats, ensuring high-quality context extraction including OCR for scanned docs.
Performance Metrics
Results & Impact
Query Response
< 30s
Semantic search returns citation-backed answers from corporate documents in under 30 seconds — down from 10+ minutes of manual hunting.
Weekly Time Saved
15 hrs
Eliminated ~15 hours of weekly manual documentation search and knowledge base maintenance per team.
Answer Accuracy
92%
Qdrant semantic retrieval with Unstructured OCR parsing achieves 92% user satisfaction on citation-backed responses.
KB Sync
Real-time
OneDrive Syncer service detects file additions, edits, and deletions via MS Graph API and re-vectorizes automatically.
File Formats
4+
Parses PDFs, DOCX, PPTX, and scanned documents with OCR via Unstructured.io for maximum knowledge coverage.
Cost Tracking
Per-User
Token usage router tracks LLM consumption per user and per model — giving full API cost visibility and budget control.