AI Prioritization Engine
An enterprise feature-request and roadmap platform that uses semantic AI to turn workforce-scale feedback into prioritized, de-duplicated, well-formed delivery work.
Role: Sole designer and builder, end-to-end
The problem
When an organization invites its entire workforce to request features and improvements, the result is a flood of overlapping, inconsistently written submissions. Conventional intake tools rely on manual tagging or simple up-or-down voting, which means duplicates phrased differently are never merged, vague requests stay vague, and the people deciding what to build can't see the real pattern behind hundreds of individual asks. The signal is buried in the volume.
What I built
I designed and built, end-to-end, a platform through which an organization's workforce submits, votes on, browses, and prioritizes feature requests, with semantic AI doing the heavy lifting throughout:
flowchart TD S(["Workforce submits a request"]) --> E["Titan Embeddings
convert meaning to a vector"] E --> D{"Duplicate of an
existing request?"} D -->|similar vector| M["Merge / link to existing"] D -->|new| C["Claude
rewrite · categorize · generate user story"] M --> RT["Route to roadmap
& delivery owner"] C --> RT
Semantic understanding, not keyword matching, drives dedup, structuring, and routing.
- Duplicate detection by meaning, not keywords. Each request is converted into a vector embedding (Amazon Titan Embeddings) that captures its meaning; comparing those vectors surfaces duplicates even when they share no words.
- Request rewriting and user-story generation. A large language model (Amazon Bedrock with Claude) clarifies submissions and aggregates patterns across many similar requests into higher-level user stories.
- Automatic categorization and routing of each request to the right delivery owners and roadmap.
- A personal dashboard and browse-and-vote interface so every contributor can see their requests, their votes, what's trending, and what shipped.
I built every layer: the React / TypeScript front end, the serverless back end (AWS Lambda, DynamoDB, S3, API Gateway), and the AI integration.
Why it matters
This is a replicable pattern for how a single organization can absorb feedback from its entire workforce and convert it into structured, prioritized delivery work using semantic AI rather than manual triage. Any organization delivering software or AI capabilities to a large user base faces the same signal-in-the-noise problem; the approach generalizes well beyond where it was built.