Getting Started
Quick start guide
Three steps from sign-in to your first answer.Sign in with your organization's Azure AD credentials. If your admin has already provisioned your account, you will land directly in the workspace. If not, contact your organization administrator to request access.Connect a file source. Carmkn supports Egnyte, Microsoft OneDrive, and SharePoint out of the box. You can also upload files directly from your local machine. Once connected, Carmkn begins processing your documents automatically.Ask your first question. Type a natural-language question in the chat interface and Carmkn will retrieve answers from your project data. The system supports 74+ file formats including Revit models, PDFs, specifications, and code files — so most of what your team already produces is ready to go.Authentication and account setup
Enterprise-grade identity management through Azure AD B2C.All authentication is handled by Microsoft Azure Active Directory B2C. Users sign in with their organization email and password. Multi-factor authentication is supported and can be enforced at the organization level by your admin.Each user gets their own account tied to their organization. There is no credential sharing — every session is individually authenticated and scoped to the user's permissions. Sessions are managed with signed JSON Web Tokens (JWT) that expire automatically.Organization-level isolation means your data is invisible to users outside your org. This boundary is enforced at the database level, not just the application layer — defense-in-depth means multiple independent controls protect your data at every tier of the stack.Connecting Egnyte
Read-only OAuth integration with folder-level access control.Navigate to Connectors in your workspace settings and select Egnyte. You will be redirected to Egnyte to sign in and authorize read-only access. After authorization, select which folder paths Carmkn should process — only files within those paths will be accessible.Files sync automatically once connected. For full details on permissions, data handling, and disconnection, see the Egnyte integration section below.Connecting Microsoft OneDrive / SharePoint
Azure AD consent flow with scoped, read-only permissions.Navigate to Connectors in your workspace settings and select Microsoft. Your Azure AD administrator must grant admin consent for Files.Read and Sites.Read.All (read-only). After consent, select which folders or document libraries to index.Both personal OneDrive and SharePoint document libraries are supported. For full details on permissions, data handling, and disconnection, see the Microsoft integration section below.Uploading your first Revit model
From .rvt file to searchable knowledge graph in minutes.Upload a Revit model (.rvt) through the file upload interface. Carmkn sends the model to Autodesk Platform Services (APS) for geometry extraction. APS converts the model into a format that Carmkn can parse — extracting spatial coordinates, element properties, and system relationships.Processing typically takes 3 to 5 minutes per model, depending on complexity. During this time, Carmkn extracts elements across every Revit category in the model: electrical panels, circuits, mechanical equipment, rooms, spaces, structural members, and more. Each element becomes a node in the knowledge graph with its full property set intact.If your project uses linked models (e.g., separate architectural, structural, and MEP files that reference each other), upload them together as a ZIP archive. Carmkn resolves the links between models and builds a unified knowledge graph that spans all disciplines.Once processing completes, you can immediately query the model. Ask about specific equipment, spatial relationships, circuit loads, or anything else captured in the Revit data.How Carmkn Works
The knowledge graph
Every document becomes a searchable network of entities and relationships.When Carmkn processes a file, it does not just store text. It extracts structured entities — panels, circuits, rooms, equipment, specification sections, code references — and creates nodes in a knowledge graph. Each node carries its source metadata, properties, and confidence scores.Edges between nodes represent relationships: a panel feeds a circuit, a circuit serves a piece of equipment, a specification references an NEC article. These relationships are typed and directional, enabling multi-hop reasoning across your project data.Cross-document linking happens automatically. When two documents reference the same equipment tag, room number, or specification section, Carmkn connects them in the graph. This means a question about a specific panel can pull information from the Revit model, the equipment schedule PDF, and the electrical specification simultaneously.The graph grows with every file you process. Each new document adds nodes and edges, strengthening existing connections and creating new ones. Over time, the graph becomes a comprehensive map of your project's technical relationships.How retrieval works
Hybrid search combining vector similarity, graph traversal, and keyword matching.When you ask a question, Carmkn runs three retrieval strategies in parallel. Vector similarity search uses high-dimensional embeddings to find document chunks that are semantically close to your query. Knowledge graph traversal follows entity relationships to find connected information. Keyword matching catches exact terms that semantic search might miss.Results from all three strategies are merged and ranked by relevance and confidence. The system considers how well each chunk matches your question, how reliable that chunk has been in past answers, and how strongly it connects to other relevant entities in the graph.Critically, Carmkn retrieves from your project data — not from generic training data or the internet. Every answer is grounded in documents your organization has uploaded and processed. If the system cannot find relevant information in your data, it tells you rather than guessing.Deterministic calculations vs AI reasoning
The AI decides what to calculate. The math is always exact.Carmkn separates AI reasoning from engineering calculations. When you ask a question that requires a calculation — load analysis, voltage drop, conductor sizing, fault current — the AI identifies what needs to be computed and gathers the inputs from your project data. The actual computation is performed by deterministic code, not by the language model.This separation matters because language models can make arithmetic errors, especially with multi-step calculations. Carmkn's calculation engines use verified formulas with NEC article citations. Every result traces back to a specific code section, table, or equation.NEC calculations cover Article 220 (load calculations), Chapter 9 Table 9 (voltage drop), Table 310.16 (ampacity), and fault current analysis. Results cite the specific articles and tables used, so a licensed Professional Engineer can verify every value before it appears in stamped documents.How the system learns
Retrieval paths that produce correct answers are strengthened automatically.Carmkn uses a synaptic learning rule to improve retrieval over time. When a retrieval path leads to a correct answer, the connection weights along that path are strengthened. When a path leads to an incorrect or low-quality answer, those weights are weakened.Reputation scores track which document chunks are reliable sources for which types of questions. A chunk that consistently contributes to accurate answers about electrical load calculations earns a higher reputation for that topic. These scores influence future retrieval ranking.Learning is scoped to your organization. Your project data improves your results — it never leaks to other tenants. Successful retrieval patterns are consolidated into reusable procedures that apply to future questions within your org, so the system gets more accurate with every interaction.File Processing
BIM model ingestion
Full property extraction across 81 Revit categories with spatial coordinate preservation.Revit models (.rvt) are uploaded to Autodesk Platform Services for conversion and property extraction. Carmkn extracts element properties across 81 Revit categories — from electrical panels and circuits to mechanical equipment, structural members, and architectural elements. Each element's full parameter set is preserved.Spatial coordinates are extracted and stored, enabling geometry-aware queries. When you ask "which panels are on the second floor" or "what equipment is in Mechanical Room 201," Carmkn uses actual coordinate data from the model — not text matching against room names.System relationships — panel-to-circuit, circuit-to-equipment, equipment-to-space — are mapped into the knowledge graph as typed edges. This enables multi-hop queries: "What is the total connected load on Panel LP-2A including all downstream branch circuits?"IFC files are processed directly without the Autodesk conversion step. The same entity extraction and knowledge graph indexing applies. All extracted data feeds into the same unified graph regardless of the source format.PDF and document processing
Multi-layer extraction for text, tables, and structured engineering content.PDFs pass through multiple extraction layers. Text extraction captures the document's content and structure. Table extraction — powered by Adobe PDF Services for complex tabular data — preserves row and column relationships that simple text extraction would flatten.Entity recognition identifies equipment tags, specification references, NEC articles, room numbers, and other structured data within the extracted text. These entities become nodes in the knowledge graph, linked to their source document and page number.The pipeline handles the full range of construction documents: equipment schedules, panel schedules, electrical specifications, engineering calculations, inspection reports, submittals, and RFIs. Each document type benefits from the same entity extraction and cross-document linking.Code and specification parsing
Structural analysis for code files and CSI-formatted construction specifications.Code files are parsed for structure: functions, classes, imports, exports, and technical references. This structural awareness means Carmkn can answer questions about code organization, dependencies, and specific implementations — not just keyword matches against raw text.Construction specifications in CSI MasterFormat are decomposed into their hierarchical structure: divisions, sections, parts, and individual requirements. Cross-references between specification sections are captured as graph edges, so a question about a material requirement can trace back through the full specification chain.Configuration files (YAML, JSON, TOML, INI) are parsed into their key-value structures. This is particularly useful for projects with complex configuration sets where understanding relationships between settings matters.Entity extraction and knowledge graph indexing
Every file produces structured entities that connect into the unified knowledge graph.Entity extraction is the bridge between raw file content and the knowledge graph. When a file is processed, the extraction pipeline identifies structured entities: equipment (panels, transformers, switchgear), circuits (feeders, branch circuits), spaces (rooms, floors, areas), specifications (sections, requirements), and references (NEC articles, standards).Each entity becomes a node in the knowledge graph with a defined type, a set of properties, and a link back to its source document and location within that document. Typed edges connect entities to each other: a panel feeds circuits, circuits serve equipment, equipment occupies rooms.Cross-document linking happens automatically. When two files reference the same equipment tag or specification section, the graph engine merges the references into a single node with multiple source links. This deduplication means a query returns consolidated information from all relevant documents, not fragmented results from individual files.Electrical Engineering
NEC load calculations (Article 220)
Deterministic load calculations using actual circuit data from your BIM model.Load calculations follow NEC Article 220 procedures using actual connected load data extracted from your Revit model. Carmkn pulls circuit amperage, voltage, and phase data from BIM elements — not from manual input or estimates.Demand factors are applied per NEC Article 220.42 (lighting), 220.44 (receptacles), and 220.50-220.56 (appliances and equipment). Continuous load multipliers (125% per NEC 210.20) are applied where applicable. Each factor references the specific NEC article that governs it.Results include a full breakdown: connected load, demand load, demand factor applied, and the NEC article citation for each calculation step. This audit trail lets a licensed Professional Engineer verify every value.All calculations should be reviewed and verified by a licensed PE before use in stamped documents or permit submittals.Voltage drop analysis
Conductor impedance calculations using NEC Chapter 9, Table 9 values.Voltage drop is calculated using conductor impedance values from NEC Chapter 9, Table 9. The calculation considers conductor material (copper or aluminum), conductor size, conduit type (PVC, steel, aluminum), and power factor.Circuit length is pulled from your BIM spatial data when available — Carmkn measures the actual routed distance between source and load, not a straight-line estimate. For circuits without spatial data, you can provide manual distances.Results are reported as both absolute voltage drop and percentage of source voltage, checked against the NEC recommendation of 3% for branch circuits and 5% for the total (feeder plus branch circuit). Calculations include the specific Table 9 impedance values used.Conductor sizing
Ampacity ratings from NEC Table 310.16 with correction and adjustment factors.Conductor sizing references NEC Table 310.16 for allowable ampacity based on conductor type, insulation rating, and installation conditions. The baseline ampacity is then adjusted for ambient temperature (NEC Table 310.15(B)(1)) and conduit fill (NEC Table 310.15(C)(1)).Continuous load factors (125% per NEC 210.20) are applied before comparing against ampacity ratings. The system recommends the minimum conductor size that satisfies all applicable derating conditions for your specific installation.When your BIM model includes conduit schedules, Carmkn factors in the actual number of current-carrying conductors per conduit for accurate fill adjustment. Results cite the specific NEC tables and correction factors applied.Power flow and fault current analysis
Distribution analysis from utility service through every level of your electrical system.Power flow analysis traces the distribution path from the utility service entrance through transformers, switchgear, distribution panels, and branch circuits. At each node in the distribution tree, Carmkn calculates loading, voltage levels, and available capacity.Fault current calculations determine the available short-circuit current at each point in the distribution system. These calculations use actual impedance values from your equipment data — transformer impedance, conductor impedance, and utility available fault current.Results identify potential issues: overloaded feeders, panels approaching capacity, and points where available fault current exceeds equipment interrupting ratings. Each calculation references the specific equipment data and impedance values used.Title 24 / CBECC compliance
California energy code compliance analysis across prescriptive and performance paths.Carmkn supports Title 24 Part 6 (California Building Energy Efficiency Standards) compliance analysis using multiple compliance methods: prescriptive, performance-based, and CBECC modeling. The system cross-references your building data against code requirements to identify compliance gaps.For prescriptive compliance, Carmkn checks envelope performance (insulation, fenestration, air barrier), HVAC efficiency requirements, lighting power density limits, and electrical power requirements against the applicable climate zone standards.Performance-based analysis compares your proposed building design against the standard reference design. Carmkn pulls envelope, HVAC, and lighting data from your project files and evaluates whether the proposed design meets or exceeds baseline performance.All compliance results reference specific Title 24 sections and standards. Results should be reviewed by a qualified energy consultant or licensed professional before submission to the building department.Time-series load analysis
Load pattern analysis for demand management, generator sizing, and capacity planning.Time-series load analysis examines how electrical loads vary across operating conditions — peak demand, normal operation, standby, and emergency scenarios. This analysis is essential for sizing generators, UPS systems, and evaluating demand response strategies.Carmkn uses your project's circuit and equipment data to model load profiles at the panel and system level. The analysis identifies peak demand windows, calculates load diversity factors, and projects capacity utilization across your distribution system.Results inform critical sizing decisions: emergency generator capacity, UPS runtime requirements, transformer loading margins, and utility demand charges. Load profile data can be used to evaluate the economic impact of demand management strategies.Enterprise & Security
Data isolation and tenant security
Physically separated databases with Row-Level Security across every table.Carmkn isolates tenant data at the database level using physically separated databases with independent credentials. This means a vulnerability in one layer of the system physically cannot reach data in another layer — isolation is architectural, not just policy.Row-Level Security (RLS) is enforced on every table that stores tenant-scoped data. Every query is automatically filtered by your organization's identifier. RLS is built into the database layer and cannot be disabled by application code.This defense-in-depth approach means that even in a worst-case scenario — a complete application-layer compromise — your data remains protected by database-level controls that the application cannot override.Immutable audit trail
Every data access operation is logged with user, timestamp, and action type.All significant operations are logged: file uploads, AI queries, document access, administrative actions, authentication events, and permission changes. Each log entry captures the user, timestamp, action type, and the resources involved.Audit logs are retained in hot storage for immediate queries and archived in cold storage for long-term compliance. Logs capture access events and metadata without storing the content of documents or AI responses.The audit trail is append-only. Log entries cannot be modified or deleted through the application — they are written to a separate storage path with independent access controls. This immutability is critical for compliance and incident investigation.SOC 2 readiness and controls
Controls aligned to all five SOC 2 Trust Services Criteria.Carmkn's security controls are designed to satisfy the SOC 2 Trust Services Criteria across all five categories: Security (CC6), Availability (A1), Processing Integrity (PI1), Confidentiality (C1), and Privacy.Key controls include: encrypted data at rest and in transit, multi-factor authentication, role-based access control, immutable audit logging, automated vulnerability scanning, and documented change management procedures. All infrastructure is managed through version-controlled infrastructure-as-code.Penetration testing is planned for Q3 2026, conducted by an independent third-party firm. An annual policy review cycle ensures that security controls evolve with the threat landscape and regulatory requirements.Connector governance and admin policies
Organization administrators control which connectors are active and what data is accessible.Organization administrators manage file connector settings through the admin interface. Administrators control which connectors (Egnyte, OneDrive, SharePoint) are enabled for their organization and which folder paths users can access.Granted paths are enforced at the API level. A user cannot access files outside the paths their administrator has authorized, regardless of what the user's file source permissions allow. This provides an additional layer of access control on top of the source system's own permissions.All administrative actions — enabling or disabling connectors, modifying granted paths, adding or removing users — are logged in the audit trail. This gives administrators a clear record of who changed what and when.Incident response
Severity-tiered response plan with defined escalation paths and post-incident review.Carmkn maintains a documented incident response plan with four severity tiers. P0 incidents (data breach, complete service outage) are acknowledged within 30 minutes, 24/7. P1 incidents (partial outage, degraded service) are acknowledged within 1 hour, 24/7. P2 and P3 incidents follow standard business-hours support timelines.Each severity tier has defined escalation paths, communication templates, and resolution procedures. The incident commander coordinates response activities, communicates status updates to affected customers, and ensures root cause identification.Post-incident reports are delivered within 5 business days of resolution. Reports include a timeline of events, root cause analysis, impact assessment, and remediation steps taken. Systemic issues identified during post-incident review are tracked through the corrective action process.Business continuity and disaster recovery
Documented recovery objectives with Azure-managed geo-redundant backups.Carmkn maintains documented recovery objectives for all critical services. Azure-managed continuous backups provide multi-week retention with geo-redundancy. Point-in-time restore is available to fine granularity within the retention window. Specific RTO and RPO targets are provided in the Service Level Agreement.The disaster recovery plan covers infrastructure failure, data corruption, and regional Azure outages. Failover procedures are documented for each critical service: databases, application servers, file storage, and the AI processing pipeline.DR test procedures are scheduled quarterly. Each test validates recovery procedures, measures actual RTO/RPO against targets, and identifies gaps in the recovery process. Test results are documented and gaps are tracked through the corrective action process.Integrations
Egnyte integration
OAuth 2.0 authorization with domain-scoped API access and folder-level grants.The Egnyte integration uses OAuth 2.0 authorization with your Egnyte domain. When a user initiates the connection, they are redirected to Egnyte's authorization page where they sign in and approve Carmkn's access request. Carmkn requests read-only permissions only.Access is scoped to specific folder paths. After authorization, administrators select which folders Carmkn should process. The folder grant is enforced at the API level — Carmkn's service account can only read files within the explicitly granted paths.Files from Egnyte are staged temporarily during processing with a 2-day time-to-live (TTL). After processing, the staged copy is removed. Extracted embeddings and metadata persist in Carmkn's database. Original files are never modified or retained long-term.Carmkn never writes back to your Egnyte store. The integration is strictly one-directional: read files, process them, store extracted data locally. Your Egnyte file structure, permissions, and content remain unchanged.Microsoft OneDrive / SharePoint integration
Azure AD OAuth with admin consent, scoped to read-only file access.The Microsoft integration uses OAuth 2.0 with Azure AD. An administrator must grant organization-wide consent for two permissions: Files.Read (read user files) and Sites.Read.All (read SharePoint site content). No write permissions are requested.Both personal OneDrive storage and SharePoint document libraries are supported through the same integration. After consent is granted, users select which folders or document libraries Carmkn should index. The same processing pipeline applies regardless of the Microsoft source.File handling follows the same pattern as the Egnyte integration: files are staged temporarily during processing, extracted data persists in Carmkn's infrastructure, and staged copies are removed after processing. Carmkn never writes data back to your Microsoft account.The integration can be disconnected at any time by revoking the OAuth consent in your Azure AD portal. Once revoked, Carmkn stops accessing new files immediately. Previously extracted embeddings and metadata remain in Carmkn's database until the organization administrator explicitly requests deletion.Autodesk Revit model processing
Full geometry and property extraction via Autodesk Platform Services.Revit models (.rvt) are uploaded to Autodesk Platform Services (APS) for conversion. APS extracts the full model geometry, spatial coordinates, and element property sets. Carmkn then processes the extracted data to build knowledge graph nodes for each element.The extraction spans the full Revit category set: electrical distribution (panels, circuits, switchgear), mechanical equipment (chillers, AHUs, pumps), plumbing fixtures, structural members, architectural elements (rooms, doors, windows, walls), and more. Each element retains its full Revit parameter set — type parameters, instance parameters, and calculated values.Spatial relationships are preserved: which elements occupy which rooms, how systems connect across floors, and where equipment is physically located. This spatial awareness enables geometry-based queries that go beyond simple property lookups.For projects with linked Revit models, upload the full set as a ZIP archive. Carmkn resolves cross-model references and builds a unified knowledge graph that spans all linked files. Element references between the host model and linked models are maintained as graph edges.Data handling and residency
All data processed and stored in Azure US West 2. AI queries carry zero data retention.All Carmkn infrastructure runs in Microsoft Azure US West 2 (Washington state). Databases, application servers, file storage, and processing pipelines are all located in this region. No data leaves the US West 2 region during normal operation.AI queries are processed through Anthropic's Claude API with zero data retention. Anthropic does not store, log, or train on any data sent through the API. Your project data is used only to generate the response to your query — it is not retained by the AI provider after the response is delivered.File connectors (Egnyte, OneDrive, SharePoint) read from your existing storage systems. Carmkn stores only the extracted embeddings, metadata, and knowledge graph entities — not copies of your original files. Staged file copies used during processing are automatically deleted after a 2-day TTL.Organization administrators can request full data deletion at any time. Deletion removes all embeddings, metadata, knowledge graph entities, audit logs, and account data associated with the organization. Deletion is permanent and non-recoverable.