Skip to content

CHANGELOG.md

Latest commit

 

History

History
72 lines (50 loc) · 4.14 KB

CHANGELOG.md

File metadata and controls

72 lines (50 loc) · 4.14 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

1.1.3 - 2026-03-05

Added

  • Case Study Link — Added the link to the devfolio portfolio page in the README.md
  • 100% Test Coverage — Added missing tests for invalid JSON edge cases and disabled LangSmith tracing

1.1.2 - 2026-03-05

Changed

  • Update classifier routing label from "Gemini 2.0 Flash" to "Gemini 2.5 Flash"
  • Update MetricsPanel test fixtures to use current model names (gpt-4.1, gpt-4.1-mini, gemini-2.5-flash, claude-sonnet-4-20250514)

1.1.1 - 2026-03-05

Fixed

  • Dashboard Scroll — Fix flex overflow in ThoughtStream with min-h-0, pin footer outside scroll area with shrink-0
  • Ticket History A11y — Replace <button> with <div role="button"> and add keyboard handler for Enter/Space
  • MetricsPanel Cleanup — Remove redundant per-model detail bars, keep provider-level bar only

Changed

  • Restructure dashboard 3-column layout with customer disambiguator, fixed textarea/submit at bottom
  • Footer now reads version dynamically from package.json instead of hardcoded string
  • Simplify footer by moving thread ID to a tooltip

1.1.0 - 2026-03-05

Added

  • Already-Resolved Detection — Resolver agent now pre-checks billing data for existing refund/credit records, skipping unnecessary LLM calls when issues are already resolved
  • Robust JSON Parsing — Action proposal parser now includes regex fallback to extract JSON from markdown-fenced LLM responses (\``json ... ````)
  • dotenv Loading — Backend config now loads .env files automatically for local (non-Docker) development
  • Redis URL Documentation — .env.example now documents password-authenticated Redis URL format

Changed

  • Resolver tests expanded with 3 new test cases covering pre-check, no-refund fallback, and fenced-JSON parsing

1.0.0 - 2026-03-04

Added

  • Multi-Agent Architecture — 4 specialized agents (Triage, Investigator, Knowledge, Resolution) orchestrated via LangGraph
  • Human-in-the-Loop (HITL) — Agent pauses for human approval before executing destructive actions (refunds, suspensions); non-destructive actions are auto-approved
  • Dynamic Model Routing — Routes simple intents to Groq Llama-3.3 ($0.00003/req) and complex intents to GPT-4.1/Gemini ($0.008/req) with automatic fallback
  • Smart Customer Validation — Handles 8 edge cases including fuzzy name matching, typo correction, disambiguation, and account status warnings
  • Self-Healing SQL — Generates SQL from natural language with auto-retry up to 3× by feeding errors back to the LLM
  • Semantic Caching — Redis-based deduplication serves identical queries in <50ms at $0.00 cost; failures are never cached
  • Real-time ThoughtStream — Watch the agent's reasoning step-by-step via Server-Sent Events (SSE) with dual User/Dev modes
  • Observability Dashboard — Track token usage, cost per request, cache hit ratio, model distribution, and database status in real-time
  • Ticket History — Recent tickets persisted in localStorage with quick resubmission
  • Database Explorer — View live counts of customers, billing records, support tickets, and internal docs
  • LangSmith Integration — Full distributed tracing with node-level spans, LLM call tracking, and graph execution traces
  • Docker Compose — One-command deployment for backend, frontend, and Redis
  • CI/CD Pipeline — GitHub Actions with Ruff linting, pytest, ESLint, and Docker build verification
  • 100% Test Coverage — Backend (pytest + pytest-cov) and frontend (Vitest + React Testing Library) both at 100%