Enterprise-grade OpenEnv environment for AI-driven customer support triage. Features stochastic noise, multi-turn dialogue reasoning, and deterministic reward scoring for benchmarking AI agents.
-
Updated
Jul 8, 2026 - Python
Enterprise-grade OpenEnv environment for AI-driven customer support triage. Features stochastic noise, multi-turn dialogue reasoning, and deterministic reward scoring for benchmarking AI agents.
Draft, then double-check: an LLM ticket pipeline with a built-in judge.
🤖 Terminal-based AI support triage agent for HackerRank Orchestrate May 2026. Processes multi-domain tickets (HackerRank, Claude, Visa) using RAG with hybrid BM25 + vector search, LLM-based classification, and safe escalation logic. Built with Python, ChromaDB, sentence-transformers, and Ollama.
AI-assisted support triage system using classification, keyword-overlap retrieval, risk evaluation, and Claude-generated responses, with a TF-IDF + semantic Python prototype.
Support triage with reviewed memory, receipt reconciliation, and Tatum transaction evidence
Multi-domain support-triage AI agent on Claude Opus 4.7. Prompt caching + structured tool-call output + prompt-injection defense + Message Batches API. Built solo for HackerRank Orchestrate (May 2026).
Support inbox drafter on synthetic data: drafts replies from help pages and holds risky requests for a person.
Lead & support intake pipeline: FastAPI webhook → Claude triage grounded in an FTS5 knowledge base → Google Sheets/CSV CRM → weekly digest. Ships an importable n8n workflow and an offline MockLLM.
Deterministic support message triage - classify support messages into bug/feature/urgent/noise
LangGraph-powered customer support agent with LangSmith tracing, automated quality evaluations, and regression gates to catch harmful prompt or model changes.
Doubt, as a service: one HTTP call turns any text in 100+ languages into a calibrated act/review/escalate/block verdict. Stateless gateway over laya.
Support Triage Agent: safety-gated, deterministic triage of support requests — prompt-injection detection, domain/request-type classification, and escalation-or-respond routing.
Offline lab for support-ticket triage: schema, tau and an invariant gate in front of fake classifiers, with one recorded live check on synthetic data.
CaseLens synthetic support triage portfolio product
To associate your repository with the support-triage topic, visit your repo's landing page and select "manage topics."