---
title: "Project Acme — Architecture"
url: https://memory.wiki/ycd140D0
updated: 2026-05-10T15:36:24.394Z
hub: https://memory.wiki/hub/yc-demo
concept_count: 12
source: "memory.wiki"
---
# Project Acme — Architecture

> _Recreated outline. Original was lost to the SAMPLE_WELCOME race._

## Overview

Acme is a post-call sales intelligence layer. Components (high level):

```
audio in → transcription → extraction → CRM write
            (Whisper)       (Claude)     (Salesforce / HubSpot)
```

## Components

### Ingestion
- Cloud upload endpoint (POST /v1/calls)
- Webhook from Zoom / Gong / Chorus
- File-store: object storage (raw audio, retained 30d)

### Transcription
- Provider: Whisper (self-hosted) or Deepgram (managed)
- Output: speaker-diarised transcript JSON

### Extraction
- LLM: Claude with structured output (JSON schema)
- Inputs: transcript, account context, recent CRM state
- Outputs: follow_ups[], decisions[], next_steps[]

### CRM bridge
- OAuth tokens per tenant (Salesforce, HubSpot)
- Idempotent write with external ID = call_id
- Approval queue: nothing writes without human OK in v1

## Data model

- `calls(id, tenant_id, audio_url, status, transcript_id, ...)`
- `extractions(id, call_id, kind, payload, status, approver_id)`
- `integrations(tenant_id, provider, refresh_token, ...)`

## Open questions

- Streaming vs batch transcription
- Multi-tenant inference vs per-tenant model pinning
- Retention: raw audio vs transcript vs extracted-only


---

## Concepts in this document
- **mdfy** _(entity)_
  Alternative approach emphasizing author-owned memory and cross-platform sharing via URLs.
- **Salesforce** _(entity)_
  Primary CRM target for bi-directional OAuth integration and follow-up sync.
- **Mem0** _(entity)_
  AI memory platform using flat vector-based storage with top-K retrieval for memory recall per turn.
- **Hubspot** _(entity)_
  Secondary CRM target supported via same OAuth integration pattern.
- **Knowledge management** _(tag)_
  Domain focus for organizing and accessing personal information in the context of AI workflows.
- **Letta** _(entity)_
  AI memory system inheriting MemGPT's hierarchical context-window architecture with core and archival memory tiers.
- **Claude** _(entity)_
  Mentioned as alternative AI platform that cannot access ChatGPT memories.
- **ChatGPT** _(entity)_
  Example AI system that can consume mdfy hubs via URL paste.
- **URL Architecture** _(concept)_
  Unified primitive design across document, bundle, and hub scopes that enables direct AI integration.
- **User-authored memory** _(concept)_
  The proposed alternative model where users control memory content rather than relying on extraction models.
- **Build in Public** _(concept)_
  Core transparency-driven strategy for sharing development progress and decisions with an audience to build trust and community.
- **AI Tools** _(concept)_
  Built-in AI capabilities including Polish, Summary, TL;DR, Translate, and Chat for content enhancement and processing.

## Concept relations (within this doc's concepts)
- **Salesforce** alternative to **Hubspot**
- **Mem0** similar to **Letta**
- **User-authored memory** implemented via **URL Architecture**
- **mdfy** product category **Knowledge management**
- **Mem0** architectural alternative **Letta**
- **mdfy** addresses **Knowledge management**
- **mdfy** implements **URL Architecture**
- **User-authored memory** core principle of **mdfy**

_Hub canonical:_ https://memory.wiki/hub/yc-demo
_Concept digest:_ https://memory.wiki/raw/hub/yc-demo?digest=1&compact=1