EPIC TECHNOLOGY · Applied AI · Sales Enablement
Technical answers for the person in the customer meeting.
I built an Internal AI Meeting Tool at EPIC for international product sales and technical presales. It helps colleagues who do not know every engineering detail respond to customer questions using live meeting context and selected product references, then turn the conversation into actionable sales notes and follow-up drafts.
Problem
The right answer depends on the right product context.
A sales colleague may understand the customer's business but lack the implementation detail needed to explain an integration, AI capability or deployment constraint. Context is scattered across meeting audio, product materials and earlier decisions.
I designed the tool around that workflow: live speech context, relevant reference retrieval, on-demand technical explanation and a structured handoff after the meeting. The goal is better-informed human responses, not autonomous selling.
Workflow
From a live question to a useful next step.
A public functional view. Customer sources and internal implementation remain private.
Capture the question
Microphone and meeting-system audio provide live speech context. The user chooses when to request help rather than delegating a customer conversation to the model.
Retrieve the right context
Selected product and project reference materials provide grounding for the current question. Source selection matters when similar products or customers are discussed.
Suggest an answer
Sales, solutions and engineering roles shape the depth of the response. The tool proposes explanations while a person owns the final customer-facing answer.
Prepare the follow-up
Structured notes capture discovery, objections, decisions, open questions, owners and next steps. CRM-ready text and customer email drafts support the next action.
Engineering decisions
Where answer quality meets business accountability.
A suggestion is not a commitment
The assistant can explain a product, but it must not quietly promise pricing, delivery dates or technical capabilities on behalf of EPIC.
Context must belong to this conversation
A fluent answer grounded in the wrong project is worse than an explicit knowledge gap. Source selection and meeting-state boundaries are central to the design.
Capture the commercial question, not just a transcript
A useful sales recap distinguishes pain points, buying signals, objections, decision criteria and agreed follow-ups from generic meeting minutes.
Keep the human in control
Answers are user-triggered; CRM-ready means copyable structured notes, not direct CRM write-back or autonomous customer outreach.
What I implemented
Desktop, speech, retrieval and meeting state.
The engineering covers a desktop interaction layer, live microphone and system-audio transcription, user-selected role and answer depth, retrieval from selected reference documents, multi-turn meeting context, and streaming answer assistance.
After the conversation, the workflow produces structured discovery notes, objections, buying signals, decisions, action owners, MEDDIC-style summaries, CRM-ready text and customer follow-up drafts. I worked across the TypeScript/Electron application, AI orchestration, persistence and evaluation paths.
The tool supports a human sales conversation rather than making commercial commitments. CRM-ready summaries are copyable notes, not automatic updates to a customer database; email drafts are reviewed before sending.
Related work