CivicTrace

Municipal complaint intake (such as for the Bhopal Municipal Corporation) is unstructured, plagued with duplicates, and manually managed. Staff spend significant hours doing sorting rather than resolving issues on the ground. Furthermore, critical hazards (live wires, open manholes, gas leaks) risk getting delayed in backlogged general queues.

I have built a resilient, operator-facing triage, deduplication, and zonal digest system:

AI-Powered Triage & Explainability: Automatically extracts category, urgency, department, and inline bilingual reasoning terms directly from Hindi/English complaint text. Safety Precedence Guard: An immutable urgency floor ensures life-safety risks (electrocution, open manholes, fire, gas leaks) are locked at CRITICAL priority with audited operator overrides. Time-Window Char N-Gram Deduplication: Clusters near-duplicate complaints across a rolling 3-day window using TF-IDF character n-grams and disjoint-set clustering without lossy misclassification blocking. Deterministic Fallback & Venue Resilience: Fully functional offline regex/keyword rules kick in automatically if the external AI API is unreachable. Built with a zero-CDN HTMX architecture to survive spotty municipal/venue Wi-Fi. Zonal Officer Weekly Digest: Instant executive reports featuring SLA tracking, safety downgrade audit logs, print CSS, and CSV injection sanitization.

Claude Code, Claude CLI, Claude Opus 5, Claude Sonnet 5