Frontend Developer· 2024–2026

MyCatalyst AI

ReactTypeScriptMulti-tenantSaaS / Data PlatformsNext.jsMUIRAGLlamaIndexContext APIOpenAI

Overview

MyCatalyst is Content Catalyst's AI-powered discovery product: a chat interface, available from any portal page, that lets subscribers ask questions in plain language and get answers drawn from the platform's licensed research (PDFs, PowerPoint, Word), with source citations and publication dates back to the original document. I built and designed the product's frontend, and contributed bug fixes, enhancements, and integration work to its backend RAG pipeline (LlamaIndex and OpenAI over a multi-tenant document store).

MyCatalyst content intelligently connected — Licensed content, connected: ask a question, get an answer with sources.

Licensing-aware by design

The hard constraint on this product isn't the chat UI, it's that answers can never leak content a subscriber isn't licensed to see. MyCatalyst enforces the same per-user permissions as the rest of the platform at retrieval time, and the underlying content is never used to train a model: RAG pulls only from what's licensed, at query time, per request.

MyCatalyst respecting licensing boundaries per subscriber — Answers stay inside each subscriber's own licensed content.

MyCatalyst source attribution and result integrity — Every answer cites its source document and publication date.

Portal-native, tenant-configurable

MyCatalyst deploys directly into a tenant's existing portal rather than as a separate destination, carrying that tenant's own branding.

MyCatalyst chat interface embedded in a subscriber portal — Embedded in the portal, matching the tenant's own branding.

I wired, designed, and built the detailed settings admins use to configure it, across three areas: General (language model selection and temperature/creativity, and how product suggestions are generated), Prompt Engineering (custom prompt tuning), and Disclaimer Management (the assistant's disclaimer text shown to subscribers). All of it without engineering involvement once it ships, a tenant can retune the assistant's behaviour themselves.

MyCatalyst settings General tab, showing language model and temperature configuration — Settings: language model, temperature, and suggested-product configuration.

MyCatalyst admin settings screen — Admin settings, tenant-side.

MyCatalyst built for analyst research workflows — Built for how analysts already work, not a separate tool to learn.

Usage analytics

I also built the usage analytics side of MyCatalyst: an admin dashboard wired end to end with the backend, giving tenants visibility into how their subscribers actually use the assistant. It tracks sessions, active users, questions and answers, average questions per session, report citations, and user feedback (likes/dislikes on answers), each filterable by date range, with a weekly usage trend chart and a most-cited-reports table.

MyCatalyst Usage dashboard overview with session, question, and citation stats — Usage overview: sessions, active users, questions, citations, and feedback.

MyCatalyst usage trend chart over time with a most-cited-reports table — Weekly usage trend, broken down by sessions, citations, and active users.

Most cited reports table, sortable and searchable — Most cited reports: which content the assistant actually draws on.

A second view clusters the raw questions subscribers asked into named topics over a chosen date range, generated on demand, which is a much faster way for a tenant to see what their subscribers are actually curious about than reading through a log of raw questions.

Question topics clustering view, grouping subscriber questions into ranked topics — Question topics: raw questions clustered into ranked, named topics.