WhyBase

• 

  WhyBase Cover

Inspiration

Every engineer knows the feeling: you're sitting in a meeting or replying to a Slack message, and someone asks "Why was this decision made?" or "Who approved this?" and suddenly everyone's scrolling through hundreds of messages, digging through GitHub issues, or worse, pointing fingers.That was my reality. Decisions were made across multiple platforms like Slack, GitHub, WhatsApp, client calls, but never in one place. The context was always scattered, always incomplete. And as engineers, we often became the scapegoat for decisions we didn't even make, simply because no one could find the original discussion fast enough. That frustration became the seed for WhyBase: what if every decision your team ever made was automatically captured, understood, and instantly retrievable?

What it does

WhyBase is an AI-powered decision memory system for teams. It connects to your existing collaboration platforms (currently Slack and GitHub) and uses Amazon Nova to automatically extract decisions from conversations, who decided it, why, and in what context. Teams can search their decision history semantically, visualize a timeline of decisions, and never again lose the why behind what was built.

How we built it

WhyBase is built on a modern full-stack architecture:

• Amazon Nova Pro (us.amazon.nova-pro-v1:0) via Amazon Bedrock for intelligent decision extraction from raw conversations
• Amazon Nova Multimodal Embeddings for semantic search across the decision knowledge base
• Slack integration to ingest team conversations in real-time
• GitHub integration to capture decisions from issues, PRs, and comments
• Tanstack Start frontend with a clean timeline and search interface
• PostgreSQL as the persistent decision store
• NestJS backend orchestrating ingestion, extraction, and retrieval

The core flow is: message ingested → Nova Pro extracts structured decision data (what, who, why, when) → embedded via Nova Multimodal Embeddings → stored and indexed → queryable instantly via natural language search.

Challenges we ran into

One of the hardest problems was deciding what counts as a decision. Not every Slack message or GitHub comment is a decision, some are questions, some are noise, some are partial agreements. Teaching Nova to distinguish signal from noise required careful prompt engineering and iterative refinement. Another challenge was cross-platform context fragmentation. The same decision is often discussed partially on Slack and finalized on GitHub, stitching those fragments into a single coherent record required building a deduplication and linking layer that was non-trivial. Latency during ingestion was also a concern, especially for Slack workspaces with high message volume.

Accomplishments that we're proud of

• Successfully extracting structured, meaningful decisions from unstructured team conversations with high accuracy using Amazon Nova Pro
• Building a genuinely useful semantic search experience powered by Nova Multimodal Embeddings — finding a decision by describing it in natural language just works
• Creating a timeline view that makes team decision history feel like a living document, not a graveyard of chat logs
• Keeping the integration lightweight, teams don't need to change how they work, WhyBase observes and learns in the background

What we learned

Building WhyBase taught us that the hardest part of AI applications isn't the model, it's the data pipeline. Getting clean, structured, deduplicated input into the model is where most of the real engineering lives. We also learned how powerful Amazon Nova's reasoning capabilities are for tasks that require nuanced judgment, like identifying intent and ownership within a conversation thread. The quality of extraction improved dramatically compared to simpler approaches.

What's next for WhyBase

The roadmap is clear: more platforms, deeper intelligence.

• WhatsApp integration, because in markets like Southeast Asia, most client decisions happen there
• Discord integration, for developer-first teams and open source projects
• Role-based access, so sensitive client decisions are visible only to the right people

The vision is simple: every team deserves a memory. WhyBase is that memory.

Built With

• docker
• lightsail
• mastra
• nestjs
• pgvector
• postgresql
• react
• typescript