06 / SOFTWARE · SAAS / SALES ENABLEMENT

An invisible second brain on every sales call, answering from the company's own product material before the rep even finishes hearing the question.

Client: Bangalore-based product-first SaaS startup

Tags: B2B SaaS · Sales Enablement · Real-Time RAG · Live Call Assistant

// Context

Reps selling a fast-moving product were expected to have instant, accurate answers to detailed spec, pricing, and integration questions — the kind of thing that lives scattered across product sheets, pricing docs, and internal battlecards, not in anyone's memory. A wrong or hedged answer mid-call cost credibility; digging for the right document while a prospect waited cost momentum.

// The Contribution

Built a meeting bot that joins the call and works entirely on the rep's side of the conversation. Before the call, it's primed with a RAG index built from the company's own product sheets, pricing material, and battlecards. During the call, it listens live, detects when the client has asked something answerable from that material, retrieves the relevant passages, and surfaces a short, source-grounded suggestion through a hoverable overlay visible only to the rep — invisible to the client on the other end of the call, whether they're on screen share or not.

// Technical Details

• Meeting bot joins the call via the video platform's own bot/API layer, listening on the rep's session
• Pre-call RAG ingestion: product sheets, pricing docs, case studies, and battlecards chunked and embedded per team
• Streaming speech-to-text plus a lightweight question-detection layer identifying retrievable client utterances in real time
• RAG retrieval against a team-scoped Postgres/pgvector store, ranked by relevance and recency
• LLM-composed, source-grounded suggestions sized for a glance, not a document dump
• Overlay rendered client-side only, excluded from screen-share capture — never visible to the other party
• Sub-2-second target latency from spoken question to surfaced suggestion
• Encrypted Postgres storage with team-scoped document isolation and API-key authentication