Problem
Case study / API v1.0.3 and web v1.0.0 released prototypes
Cricket Intelligence System
A dataset-grounded cricket question system with deterministic routing and honest degraded behavior.
Target user
Cricket viewers exploring player, team, match, comparison, and leaderboard questions.
Solution
A React client and Express API classify questions, normalize entities, compute typed dataset responses, and preserve an explicit unsupported/degraded path.
Personal contribution
Built the routing and response contracts, dataset paths, web integration, regression cases, startup/status checks, and degraded-retrieval fix.
Architecture
Boundaries before buzzwords.
- React query client
- Express routing and typed responses
- Deterministic dataset services
- Optional local Chroma helper
Main workflow
State moves with a reason.
- 01
Normalize the question and entities
- 02
Select a deterministic action
- 03
Compute from available data
- 04
Attach typed evidence
- 05
Return unsupported/degraded guidance when necessary
Technical decisions
The trade-offs that define the system.
- Statistics come from deterministic data paths
- Optional retrieval cannot silently downgrade typed team responses
- Unsupported questions remain explicit
Difficult engineering challenge
The regression-prone edge was preserving a frontend-compatible typed response when the optional vector archive was empty or locally broken.
Verification
Evidence and boundaries.
Testing
- 19 API routing cases
- 5 API HTTP integration tests
- 3 web Playwright workflows
- API and web production checks
Security and privacy
- No language-model-generated statistic is treated as evidence
- Public deployment would require request abuse limits
Known limitations
- Optional Chroma archive is empty in the verified environment
- Dataset scope bounds available answers
Focused future work
- Add provenance and golden statistical calculations for common queries
Verified demonstration
Rendered product behavior.
Loading video metadata…