Tideline
Turn a flood of support tickets into the one incident that matters
Enterprise AI system
Tideline
Turn a flood of support tickets into the one incident that matters
This production is a private summary: description, learnings and stack are shown, the code stays private.
Draft. Written from the program notes and pending the author’s review.
- Status
- Now showing
- Type
- Enterprise AI system
- Running since
- January 2025
- Role
- Lead engineer
- Domains
- support operations, incident management
- Cast
- Python, PostgreSQL, Kafka, OpenTelemetry
Context and problem
Support-ticket intelligence: LLM enrichment behind a privacy gate, incident grouping and idempotent delivery.
Assumption: during an outage a support team receives many near-duplicate tickets and has no way to tell which single incident is driving them, so the same problem is investigated several times in parallel.
Tideline is the user-reported layer of the Weave suite: it groups the tickets that arrive during an outage by the keys they share, surfaces the one incident behind them and publishes it as an event so remediation can start.
How it was built
Built in 3.8 h · 167 tests · 3 audits
Source: Trysac/07-tideline docs/development-report.md (lines 16, 18); devlog/ci.jsonl line 61; 229 min of commit time, dated 21 November 2026.
Architecture and interfaces
- Provides
- events outage.*
- Shared keys
- user_code, request_id, endpoint, error_code
What the project taught
- Grouping tickets by the keys they share (user code, endpoint, error code) is more reliable for outage detection than text similarity, because users describe the same failure in different words.
- A human confirmation step before an incident is declared keeps false positives tolerable and keeps the team in control of what gets escalated.
- Publishing incidents as events with a shared taxonomy lets a remediation system subscribe without coupling the two codebases.
- Agreeing on common keys (user code, request id, UTC time, endpoint, error code) early is what later makes the suite joinable; adding them afterwards is the expensive part.
Case study in preparation
The full case study adds these sections when it is published:
- Constraints
- Architecture diagrams (C4 context and containers)
- Three key decisions
- What went wrong
- Results with dated evidence
- Demo video
- What I would do differently
In the project mesh
Highlights
event driven, shared keys, human in the loop