Flagship case ยท AI Engineering Control
System Digital Twin in Four Weeks
How two people built a textual system twin for a large inherited codebase in four weeks.
Disclosure
The platform is anonymized. Four weeks is measured; six months is the projected duration of the manual audit, not an observed counterfactual.
One-sentence outcome
A two-person team built a working textual system twin in four weeks instead of a manual audit projected to take six months.
Context and stakes
The codebase combined about thirty microservices, hundreds of undocumented stored procedures, and almost no trustworthy architectural map, yet delivery decisions still had to be made inside an active business.
Baseline and constraints
A classical audit would have required weeks just to stabilize orientation, and the available window was four weeks with two people. RAG had already shown itself too brittle for the task boundary that mattered.
Exact role
Chief Technology Officer designing the discovery method, shaping the textual knowledge base, and using generated system context as a controlled way to regain engineering control.
Trigger to measured change
- Trigger
The inherited platform was too large and too undocumented for a manual architecture audit inside the available deadline.
- Constraint
The team needed a reusable picture of the system, not one more folder of disconnected notes that would drift after first use.
- Decision
Reverse-engineer the codebase and stored procedures into a git-tracked textual knowledge base, then compile that material into system context the agents could navigate progressively.
- Measured change
The resulting digital twin was finished in four weeks with two people, replacing a manual audit projected to take six months.
Built a system digital twin in four weeks with a two-person team, replacing a manual audit projected to take six months.
Keep Projected attached to the six-month comparator and keep the story inside its NDA-safe boundary. The case demonstrates inspectable system understanding, not the largest-client throughput or a universal anti-RAG claim.
- Context
- Inherited casino platform with roughly 30 microservices and 300+ stored procedures.
- Timeframe
- Four-week recovery window documented in the phase-0 ledger on 2026-09-02.
- Baseline
- Manual audit projected at six months; projected remains part of the comparator.
- Role
- Chief Technology Officer designing the discovery method, shaping the textual knowledge base, and using generated system context as a controlled way to regain engineering control.
- Source
- Review the source record
- Provenance
- Russian and English source posts plus the master profile.
- Confidence
- B - first-hand artifact story with a projected comparator.
- Disclosure
- Anonymized case; the platform and client remain unnamed.
Measured or observable result
Built a system digital twin in four weeks with a two-person team, replacing a manual audit projected to take six months. The projected comparator stays attached to the six-month line.
Attribution and caveat
Keep Projected attached to the six-month comparator and keep the story inside its NDA-safe boundary. The case demonstrates inspectable system understanding, not the largest-client throughput or a universal anti-RAG claim.
Retained capability
The organization retained a textual map, reusable summaries, and a navigation layer that made further AI-assisted engineering work legible instead of magical.
Related problem, next case, and CTA
The related case follows what happened next: engineers adopted the agent tools once daily work made them useful.