Phase 1: Deploy and Connect (Weeks 1-2)
Objective: Platform running on customer infrastructure with initial data sources connected and the Chat Agent answering questions.1
Provision Infrastructure
Your IT team provisions server resources and sets up the Docker or Kubernetes environment. See Infrastructure Requirements for specs.
2
Deploy Platform Components
Superatom deploys the Main Agent, WebSocket server, Web UI, and Report Engine on your infrastructure.
3
Connect Data Sources
Data-source agents are deployed for each connected source. Read-only database credentials are provided by your DBA team.
4
Automated Schema Analysis
The system runs automated schema analysis and statistical profiling, then generates an initial semantic model with proposed business-term mappings.
5
Validate Priority Questions
Your designated power users validate 5-10 high-priority questions against live data. SSO integration and initial role-based access control are configured.
Responsibilities
Data Source Connection Times
Deliverable: Chat Agent is live. Power users can ask questions and receive visual answers from connected data sources.
Phase 2: Build Knowledge and Refine (Weeks 2-4)
Objective: The semantic model is enriched with organizational context. Accuracy improves through structured feedback and tribal knowledge capture.1
Use the Chat Agent Daily
Your team uses the Chat Agent for regular analytical work, surfacing real questions and real-world edge cases.
2
Provide Corrections and Context
When interpretations are inaccurate, users provide corrections (e.g., “Revenue here means net revenue excluding returns”). These corrections are captured as knowledge nodes or golden paths — not used for LLM retraining.
3
Create Knowledge Nodes
Domain experts define critical business terms and calculation rules as structured Knowledge Nodes at Global, User, and Query levels.
4
Configure Access Controls
Column-level and row-level access controls are set up based on your organization’s permission structure.
5
Expand Data Sources
Additional data-source agents are deployed as new sources are identified beyond the initial set.
Responsibilities
How Accuracy Improves
Accuracy improvement in this phase comes from two mechanisms:Semantic Model Refinement
Semantic Model Refinement
When users correct a misinterpretation, that correction is captured as a knowledge node or a golden path. Future queries that match the same pattern use the corrected interpretation. This is not LLM retraining — it is updating the business rules and mappings that guide query generation.
Tribal Knowledge Capture
Tribal Knowledge Capture
Domain experts explicitly define how their organization uses terms, what filters should be applied by default, and what business rules govern calculations. These definitions are stored as structured knowledge nodes and applied automatically to relevant queries.
Accuracy by Query Category
Simple lookups may reach near-perfect accuracy in week 1. Complex cross-source analysis may require additional knowledge nodes through weeks 2-4. Edge cases surface over time and are addressed as they appear.
Deliverable: Semantic model is enriched with organizational knowledge. Common questions are answered accurately. Tribal knowledge is captured and applied. Permission structure is configured.
Phase 3: Dashboards, Reports, and Workflows (Weeks 4-6)
Objective: Operational dashboards are built, reports are scheduled, automated workflows are configured, and mobile access is enabled.1
Build Dashboards
Your team builds dashboards for key use cases: executive overview, sales performance, inventory health, and other operational views.
2
Configure Scheduled Reports
Reports are set up on recurring schedules — daily KPI summaries, weekly reviews, monthly executive reports — with delivery via email, Slack, or Teams.
3
Deploy Anomaly Detection in Shadow Mode
The anomaly detection engine observes historical patterns to learn what “normal” looks like for each monitored metric. During shadow mode, it generates alerts internally but does not deliver them, allowing threshold tuning without alert fatigue.
4
Set Up Workflows and Actions
Triggered workflows (anomaly detection, threshold alerts, low-stock warnings) and approval workflows for write-back actions are configured.
5
Enable Mobile Access
Mobile app access is enabled for on-the-go analytics.
Responsibilities
Deliverable: Dashboards are live and shared. Reports run on schedule. Workflows are configured with approval gates. Anomaly detection is learning in the background. Mobile access is available.
Phase 4: Go-Live and Activation (Weeks 6-8)
Objective: Full production deployment. Autonomous monitoring is activated. All user groups are onboarded.1
Review Shadow-Mode Results
Your team reviews anomaly detection results from shadow mode and approves or tunes thresholds before going live.
2
Activate Autonomous Monitoring
Anomaly detection transitions from shadow mode to live alerting with root cause analysis included in every alert.
3
Onboard All Teams
Remaining teams and user groups are onboarded. End-to-end workflows are validated: question to insight to action to result.
4
Security Audit
Superatom conducts a security audit, verifies access controls, and validates audit trail completeness.
Responsibilities
Deliverable: Platform is fully operational. Chat, dashboards, reports, mobile, workflows, and autonomous monitoring are all live. All teams are onboarded.
Phase 5: Optimization and Expansion (Weeks 8-12)
Objective: Performance optimization, expanded use cases, stress testing, and knowledge transfer to the customer’s admin team.1
Performance Optimization
Pre-materialize common cross-source joins, optimize golden path caching, and tune LLM call frequency for faster response times.
2
Expand Use Cases
Connect additional data sources and extend to new departments and question types. Ongoing feedback refines accuracy for new query patterns.
3
Load and Stress Testing
Run load testing and stress testing at production scale to verify performance benchmarks.
4
Advanced Workflows
Configure complex multi-step processes, escalation paths, and automated decision chains.
5
Admin Training and Knowledge Transfer
Train the customer’s admin team on platform management, knowledge node administration, and data source onboarding.
Responsibilities
Deliverable: Fully optimized deployment. Performance benchmarks verified. Admin team trained. Expansion roadmap defined.