liamguest.com
    Tempus Brief
    Tempus Brief

    Tempus Brief

    icon
    Tags
    Machine LearningExplainable AITechnical CommunicationProject Management
    Date
    March 15, 2026
    Description

    GenAI sales intelligence tool built in one week as a PM case study for Tempus AI. Ingests CRM notes, market volume data, and a clinical product knowledge base to generate a ranked oncology provider list, objection handler, and personalized meeting script. Built with the Anthropic API and streaming SSE. Shipped as a single HTML file with zero dependencies.

    Link to Project
    GitHub

    Demo

    Watch the Loom demo

    About

    Tempus Brief is a GenAI-powered sales intelligence tool built in one week as a PM case study prototype for Tempus AI. The prompt: design and build a tool that helps oncology sales reps prioritize their territory and walk into every meeting prepared.

    The core insight driving the product is that oncology sales reps sit on three valuable data sources — CRM interaction history, market volume data, and clinical product knowledge — but have no unified way to turn those inputs into a pre-meeting brief. Tempus Brief closes that gap.

    What It Does

    Given a territory of oncology providers, Tempus Brief:

    • Ranks providers by impact score using a weighted model that blends clinical signal (patient volume, recent drug approvals) with relationship warmth (CRM notes). Reps can toggle between Clinical, Even, and Warmth ranking modes depending on their goals.
    • Generates an objection handler grounded in Tempus's actual product knowledge base — not generic responses, but answers tied to specific test capabilities, TAT benchmarks, and reimbursement data.
    • Writes a 30-second elevator pitch in first-person rep voice, ending with a clear ask — ready to use as-is or adapt on the fly.
    • Surfaces a "Why Now" frame tied to recent FDA drug approvals relevant to each provider's patient mix.

    Product Decisions Worth Noting

    The ranking logic treats priority score as a transparent, inspectable output — not a black box. Each provider card shows a plain-English rationale and a collapsible score breakdown so reps understand why someone ranked where they did. This was a deliberate choice: in a clinical sales context, a rep who can't explain their prioritization to a manager or themselves won't trust the tool.

    The single-file, no-backend architecture was a constraint that shaped good product thinking. Everything — mock data, knowledge base, streaming response parsing — lives in one HTML file. No setup, no login, no infrastructure. A rep could run it from a thumb drive if they had to.

    Stack

    • Anthropic API (Claude Sonnet) — streaming SSE responses
    • Vanilla HTML / CSS / JS — zero dependencies, single file
    • Mock data: CRM notes (9 providers, Boston / Chicago / Houston), market intelligence CSV, Tempus product knowledge base sourced from tempus.com

    Data Sources Ingested

    • Market Intelligence CSV — oncologist list, patient volume by cancer type, recent drug approvals
    • Tempus Product KB — full test portfolio (xT CDx, xR, xF+, xM MRD, hereditary panels), TAT benchmarks, objection responses
    • CRM Notes — prior interaction history, known objections, relationship warmth scores per provider