Ella (Effortless Learning & Lookup Assistant)

GIF
Architecture Diagram
User Persona
Challenge Statement
Before & After - New Employee
Before & After - Seniors and Experienced Team Members
Conclusion
GIF
Stage 1 - Instant answers from from past resolved conversations
GIF
Stage 2 - Unknown questions are sent to #faq channel
GIF
Stage 3 - Help is saved to knowledge base by Ella
GIF
Stage 4 - Repeated questions are auto answered by Ella

Inspiration

The inspiration for Ella (Effortless Learning & Lookup Assistant) comes from a universal challenge we've all experienced: knowledge silos in growing teams. As development teams scale and new members join, we consistently see the same pain points:

Repeated Questions: The same technical questions asked over and over by different team members, despite having been answered before
Slack/Team Chats Knowledge Graveyard: Valuable solutions, decisions, and tribal knowledge buried deep in chat threads, impossible to discover when you need them most
Onboarding Friction: New developers struggling to find answers to basic questions about processes, architecture decisions, and team conventions
Context Switching: Engineers constantly interrupting each other for information that should be easily accessible

We realized that most organizational knowledge lives in two disconnected worlds: formal documentation (often outdated) and informal team chat conversations (searchable but not discoverable). The gap between these creates frustration for everyone new hires can't find answers, and experienced team members become human search engines and thats when we decided to build Ella an AI Agent that converts chats into a living knowledge base.

What it does

Ella transforms scattered Slack conversations into an organized, AI-powered knowledge base:

🔍 /ask_ella - Instant answers from your team's knowledge
📚 /add_to_document - Smart content capture with AI relevance checking
🎯 Semantic Search - Natural language, not just keywords
📊 /document_stats - Track knowledge growth

Smart Features: AI relevance scoring, duplicate detection, anonymous questions, automatic categorization.

How we built it

Multi-Agent Architecture

QnA Agent: Handles questions using RAG-powered search
Document Agent: Manages content ingestion and curation
Orchestration: Google ADK coordinates agent collaboration

Challenges we ran into

Content Quality Control - Ensuring only valuable information enters the knowledge base while preventing spam and irrelevant content through AI-powered relevance scoring, semantic deduplication and human in the loop.
Slack threading context preservation across conversations

Accomplishments that we're proud of

🚀 End-to-end pipeline: Slack → AI → searchable knowledge base
⚡ Sub-second responses with optimized RAG retrieval
🧠 Context-aware: Captures full thread conversations, not just messages
🎯 Smart curation: AI prevents spam while preserving valuable content

What we learned

Technical: RAG architecture, Vertex AI ecosystem, async Python patterns, multi-agent orchestration, semantic similarity.

Product: Building user persona, framing the challenge and making AI feel like a natural teammate, balancing automation with human curation, the difference between search and knowledge discovery.

Team: Building production-ready AI systems rapidly, integrating multiple cloud services seamlessly.