Peak
AI-Native Operations Platform
Role: Founder & Lead Engineer | Timeline: 2024 – Present
Stack: Next.js 15, TypeScript, OpenAI Realtime API (GPT-5), Supabase/PostgreSQL, MCP Protocol, React Flow, Tailwind
Overview
Peak is a voice-first operations platform powered by F.R.I.D.A.Y. (Flexible Response Intelligent Digital Assistant Year-round)—a sophisticated AI agent that orchestrates project workflows, coordinates teams, and eliminates the friction between intent and execution. Operate at compound velocity where saying what you need becomes doing what you need.
Core Intelligence Systems
Voice-First Interface
- Natural language project management via OpenAI Realtime API with sophisticated British-accented AI personality
- Sub-second response times powered by deterministic caching layer
- Context-aware conversations spanning multiple sessions with semantic memory
- 90+ built-in actions across workspace, system, and research domains
- Long-horizon memory with automatic context recovery and follow-up tracking
F.R.I.D.A.Y. Agent Architecture
- Unified voice broker tools routing to specialized domains (projects, tasks, documents, calendar, memory, agents, research)
- Multi-agent orchestration: deploy specialized research agents in parallel for complex analysis and synthesis
- Real-time MCP (Model Context Protocol) integration for direct database operations
- Extended thinking mode for transparent reasoning on complex decisions
- Session continuity bridge: reconnects context after 30+ minute gaps with intelligent greeting system
- Follow-up memory buffer with latency-aware pacing and momentum detection
Calendar Intelligence
- Bidirectional Google Calendar sync with automatic conflict detection and resolution
- Energy-aware scheduling that places tasks at optimal times based on patterns
- Task-event linking with automatic note flow back to projects
- Meeting context awareness (Hush Mode) for brevity during active calls
- Natural language event creation: “schedule sprint planning next week” → optimal slot selected, conflicts resolved
Interactive Builder (Neural Canvas)
- Voice-to-diagram generation: “show the product roadmap visually” → dependency-aware diagrams in seconds
- 15+ intelligent layout algorithms: radial, flowchart, network hub, blueprint grid, Fibonacci spiral, golden ratio, pyramid hierarchies
- Multi-domain support: BPMN, UML, electrical engineering, mechanical systems, chemistry, business workflows
- Smart element rules with auto-spacing (min 100px), color harmony by role, criticality-based sizing
- Natural language editing: “make QA bigger” → re-weighted layout with critical path emphasis
- Technical specification inline rendering and connection labels with precise specs (e.g., “480V DC”, “CAN 1Mbps”)
Agent Orchestra
- Deploy specialized research agents with natural language: “deploy agent on Layer-2 market dynamics”
- Parallel multi-agent execution with source-linked synthesis briefs
- Agent analytics dashboard with performance tracking, activity monitoring, and configuration management
- Configurable agent types (research, analysis, monitoring) with custom scopes and priority levels
- Real-time agent status views: overview, performance, activity logs, analytics
Visual Intelligence
- Screenshot understanding and visual search across workspace
- Automatic diagram generation from task lists, workflows, and project structures
- Critical path emphasis with presentation-ready exports (JSON + Markdown)
- Support for technical blueprints, org charts, system architectures, and data pipelines
- Turn 15 tasks into a workflow diagram with dependencies and milestones in seconds
Semantic Memory & Context
- Route stack tracking (last 5 navigation states) with “back” command support
- Project context retention up to 2 hours for active discussions
- Conversation momentum detection: adapts tone based on user velocity (high momentum = concise/action-oriented)
- Smart context extraction: infers project from last 10 turns before asking
- Memory operations: load_state, get_context, find_memories, update_state with automatic redaction of PII/secrets
- Temporal greeting engine: time-aware greetings with signal modifiers (momentum, alchemy mode, weekend mode)
Technical Architecture
Real-Time Performance
- Deterministic caching with sub-second pre-computed context retrieval
- Live sync: changes propagate in real-time across users and sessions
- Cache statistics, pagination, and intelligent data source priority (MCP → cache → manual fetch)
- Global inference fabric for consistent AI performance regardless of location
- Near-instant responses: typical sub-500ms for cached queries
Multi-Tenant Security
- Row-level data isolation with automatic user_id + organization_id filtering on all MCP queries
- Hierarchical access model: Personal (user_id) → Organizational (org_id) → Universal workspace
- SSO/SAML ready, RBAC with granular permissions for actions and data
- Encrypted at rest and in transit with comprehensive audit trails
- Mandatory confirmation flows for destructive actions (delete, truncate, production deploys)
Tool Tiering System
- Configurable modes: voice (3 tools), minimal (4), standard (6), mcp (5), mcp_standard (7 – recommended)
- Dynamic tool filtering based on environment flags for optimal performance
- Voice-optimized function selection for low-latency responses
- Unified broker architecture: executeWorkspaceAction, executeSystemAction, supabaseDbTool (MCP)
Advanced Workflows
- Progressive error recovery: silent retry → notification + fallback → manual guidance → offline capabilities
- Navigation discipline: never mixes data retrieval with navigation in the same turn
- Bulk operations: multi-task creation, contact imports (max 100/batch), project templates
- Smart completion detection: waits for complete user input before action (no mid-speech interruptions)
- Tool budget: 10 calls/turn for voice mode, up to 50 for complex operations
Key Features
Project & Task Management
- Full CRUD operations with priority levels (lowest, low, medium, high, highest) and status tracking
- Hierarchical task management with natural language context extraction
- Bulk task creation: “create three tasks” triggers specialized bulk_create flow
- Progress tracking with visual bars, pinned projects, and completion analytics
- Smart project detection: checks last 10 turns for project mentions before asking
Document Intelligence
- Semantic search across all project content with query highlighting
- Document type categorization: meeting notes, specifications, reports, general
- Version tracking with “most recent” indicators and update timestamps
- Voice-optimized document reading with section summaries
Navigation System
- Intelligent routing: page → project → project_tab → task/document hierarchy
- Synonym mapping: “dashboard/home/overview” → dashboard, “tasks/board/kanban” → tasks tab
- Breadcrumb tracking with micro-confirmations: “Current: Peak Platform › Tasks”
- Index & pointer references: “open the 3rd project” resolves against last rendered list
- Back/previous navigation with route stack history (max 5 states)
Communication System
- SMS preview-confirm-send workflow via Twilio integration
- Favorite contact management with bulk import from phone/CSV sources
- Sentiment-aware messaging: warm tone for personal notes, professional for business
- Contact CRUD operations with phone number validation and relationship tracking
Dashboard & Widgets
- Voice-controlled widget expansion: weather, agents, time tracking, project summaries
- Modal drilldowns with multi-tab views (agent analytics, extended weather, air quality, maps, historical data)
- Time tracking with automatic duration logging, productivity insights, and entry history
- Integration-ready weather data with alerts, forecasts, and energy pattern recommendations
Web Research
- Real-time search with source attribution and time-stamped results
- Voice-optimized reading: summary-first, key insights highlighted, URLs omitted in speech
- Deep research mode for comprehensive analysis with structured synthesis
- Automatic “As of [timestamp]” annotations for time-sensitive content
PEAK Format
- Structured visual output for lists with priority sections (🔥 PRIORITY → ● MEDIUM → ▫ LOW → ✅ COMPLETED)
- Completion statistics headers: “4 pending • 6 completed” with total counts
- Pagination support with “next page” / “prev page” voice commands
- Progress bars with 30-character visual representation rounded to nearest percent
- Empty state templates with next-step suggestions
Differentiation
Unlike traditional project management tools or generic AI assistants, Peak combines voice-first interaction with enterprise-grade intelligence. F.R.I.D.A.Y. doesn’t just take commands—it understands context across hours and sessions, remembers decisions, detects bottlenecks before they escalate, and proposes the next best move with genuine personality and wit.
The platform forms covalent bonds with your workflow (Covalent Intelligence™), strengthening with every decision, artifact, and outcome. Say what you need; watch blockers surface in real time; ship with certainty—without changing tools or fighting interfaces.
Impact & Scale
- Near-instant AI responses: sub-500ms for cached queries, sub-2s for complex orchestration
- Supports complex multi-step workflows with up to 50 tool calls per operation
- Multi-language support with automatic dialect matching and language detection
- Adaptive personality modes: Alchemy Mode (exploratory), Hush Mode (meeting-aware brevity), Weekend Mode, Late-Night Nudges
- Conversation momentum tracking: high momentum (>60%) = concise/action-oriented, low momentum (<30%) = activation energy and suggestions
- Production deployment on Vercel with global edge distribution
Technology Leadership
Peak demonstrates advanced full-stack engineering with cutting-edge AI integration:
- First-wave adoption of OpenAI Realtime API (GPT-5 Thinking) with voice-optimized function calling
- Custom MCP (Model Context Protocol) server implementation for direct database operations
- Sophisticated agent handoff architecture: F.R.I.D.A.Y. ⇄ Interactive Builder with context preservation
- Deterministic response system with formatted/message/\__deterministic__ precedence rules
- Multi-domain tool broker with unified error handling and progressive recovery patterns
- Real-time bidirectional sync with conflict resolution and optimistic UI updates
Live: Active development with production deployment | Vision: Operate at compound velocity—where intent becomes action without friction
Built for decisive teams, designed for compound velocity, powered by F.R.I.D.A.Y.