# Deemto / Nano AI — Original Brainstorm: Technical Summary

## Context
This document summarizes the original brainstorm session that led to the Nano AI
project. It captures every strategic and technical decision made, in order, so we
can reference it as a canonical source of intent.

---

## 1. Strategic Filters (Locked In Early)

The founder established hard constraints before picking any idea:

- **No culturally sensitive domains** (hiring, performance reviews)
- **No heavily regulated domains** (healthcare, finance compliance)
- **No subjective-outcome businesses** (where success depends on human judgment)
- **Target: B2B SaaS** — companies that already spend on software
- **Outcomes must be measurable** — time saved, errors reduced, revenue increased
- **Must serve globally** — not geography- or culture-dependent
- **Must be AI-native** — not a thin wrapper on existing tools
- **Low politics** — objective, process-driven work

## 2. Idea Selection Funnel

| Stage | Decision |
|---|---|
| Category | AI for B2B workflow automation |
| Track | Knowledge Copilot (small, local models) |
| Scope | E — Horizontal Knowledge Engine (long-term) |
| Wedge | A — Operations (first vertical entry point) |
| First workflow | D — Cross-tool workflow automation |
| First use case | A — New hire onboarding |
| First integrations | Slack + Google Workspace + Jira/Asana |
| UX model | C — Hybrid (Slack for commands, dashboard for control) |
| Builder style | C — Hybrid (list view + flowchart view, AI-synced) |

## 3. Core Technical Philosophy

### Small Model Thesis
- Intelligence comes from **architecture, not parameter count**
- Target model sizes: 1B–7B parameters
- Models run **locally or on private cloud** — no data leaves the company
- **Domain-adaptive micro-models**: separate small models for Ops, Support, Engineering, HR
- Deterministic reasoning > stochastic generation for business workflows
- Architecture: **Retrieval + Reasoning + Execution** (3-layer stack)

### Three-Layer Architecture
1. **Retrieval Layer** — indexes all internal docs, SOPs, policies; fast local vector search
2. **Reasoning Layer** — small model interprets, generates steps, fills forms, classifies
3. **Execution Layer** — agents that perform actions across tools (Slack, Jira, Google, etc.)

## 4. AI Identity & Behavior Decisions

| Decision | Choice |
|---|---|
| AI identity | **The Ops Architect** — designs, optimizes, evolves operational systems |
| Intelligence style | **Hybrid Intelligence** — balances efficiency, compliance, and automation |
| Workflow generation | **Transformative** (primary) with **Interpretive** fallback |
| Communication tone | **Adaptive** — adjusts to user role, company culture, workflow sensitivity |
| Mission statement | "We architect, automate, and evolve your workflows continuously" |
| Headline | "Your AI architect for operational excellence" |
| Tagline | "Your operations, automated and continuously improved by intelligent architecture" |

## 5. Product: Three Pillars

### Pillar 1 — ARCHITECT
- SOP ingestion (PDF, Google Docs, Notion, Confluence, text)
- AI generates two workflow versions: Interpretive + Transformative
- Hybrid builder: list view ↔ flowchart view (always in sync)
- Structural optimization: merge steps, parallelize, detect dependencies
- Compliance mapping, role-specific branching
- AI suggests improvements during editing (co-designer UX)

### Pillar 2 — AUTOMATE
- Slack as command interface ("Onboard Sarah Chen as Backend Engineer starting April 12")
- Google Workspace: create accounts, add to groups, send email, schedule meetings, share docs
- Jira/Asana: create tasks, assign, set deadlines, track progress
- Cross-tool orchestration: one command triggers actions across all integrated tools
- Real-time progress tracking in dashboard

### Pillar 3 — EVOLVE
- Execution analytics: time per step, bottlenecks, delays, errors, skipped steps
- AI analyzes logs and proposes improvements autonomously
- "Step 7 is consistently delayed — move it earlier"
- "These two tasks can run in parallel"
- Accept/reject improvement suggestions in UI
- Continuous optimization loop — workflows get better over time

## 6. MVP Feature Set

### Workflow Architecture
- SOP upload → AI generates Interpretive + Transformative workflows
- Hybrid builder (list + flowchart, synced)
- Step types: human task, automated task, approval, wait, branch
- Save as reusable templates (Engineer Onboarding, Designer Onboarding, etc.)

### Execution
- Slack slash command triggers workflows
- AI extracts: name, role, start date, department
- Google Workspace: account creation, groups, calendar, docs, email
- Jira/Asana: task creation, assignment, deadline tracking
- Slack: notifications, reminders, status updates, approvals
- Real-time dashboard with execution logs

### Evolution
- Basic analytics (time per step, bottlenecks)
- AI improvement suggestions
- One-click "Apply suggestion" to update workflow

## 7. Data Model (MVP)

**Entities:** Workflow, Step, Role, Tool Action, Condition, Branch,
Execution Instance, Execution Log, Improvement Suggestion

**Key relationships:**
- Workflow → many Steps → many Tool Actions
- Step → Conditions, Branches
- Workflow → many Execution Instances → Logs
- Logs → feed Improvement Suggestions

## 8. Architecture (MVP)

- **Frontend:** React or Svelte dashboard + hybrid builder + Slack bot
- **Backend:** API layer + workflow engine (state machine) + integration layer + AI reasoning
- **Storage:** Postgres (workflows, steps, logs) + Vector store (SOP embeddings)
- **AI:** Small model (1B–7B) + retrieval pipeline + deterministic wrappers
- **Deployment:** Local-first (on-prem, private cloud, intranet)

## 9. Pricing Strategy

| Tier | Price | Features |
|---|---|---|
| Starter (SMB) | $99–$199/mo | 1 workflow, 3 integrations, Slack bot, basic analytics |
| Growth (Mid-size) | $399–$799/mo | Unlimited workflows, all integrations, advanced analytics, evolution |
| Enterprise | $2k–$5k/mo | On-prem, custom integrations, compliance, dedicated support |

## 10. Target Customer Profile

- 50–300 employee tech companies
- Growing fast, hiring regularly
- Using Slack + Google Workspace + Jira/Asana
- Ops team overwhelmed by manual processes
- No internal automation expertise

## 11. Six-Week Build Plan (Original)

| Week | Focus | Key Deliverable |
|---|---|---|
| 1 | Foundations | Slack bot, Google/Jira APIs, workflow engine skeleton, DB schema |
| 2 | SOP Ingestion + AI | Upload SOP → AI generates Interpretive + Transformative workflows |
| 3 | Hybrid Builder | List view + flowchart view (synced), drag-and-drop, step types |
| 4 | Execution Layer | Slack triggers → Google + Jira actions → real-time dashboard |
| 5 | Evolution Layer | Analytics, bottleneck detection, AI improvement suggestions |
| 6 | Polish + Demo | Clean UI, demo script, pre-built templates, first user testing |

## 12. Long-Term Vision

**Wedge:** Ops onboarding automation
**Platform:** Horizontal AI knowledge engine — the internal AI OS of the company
**Expansion path:** Ops → Support → HR → Engineering → Finance → Security → Legal
**Moat:** Small-model local-first architecture + coordination protocol + accumulated workflow intelligence
