feat: support workload-specific model configuration, trust role description grounding, and remove emojis

fix: import InstructorRetryException from instructor.core to avoid lazy-import AttributeError

feat: implement database error logging for start and turn conversation failures

fix: recalculate derived outreach fact IDs during model copy and validation to avoid stale values

feat: migrate structured model outputs to Instructor with repair recovery fallback

fix: resolve structured format fallback, nested validation repairs, and add turn fallback protection

feat: default to openai/gpt-oss-120b:free model

fix: fall back to alternative structured output mode on incomplete model responses

feat: track candidate profile in dialogue turns

fix: serialize UUIDs to strings in database interactions payload

feat: trace agent configuration in database

feat: integrated database

fix: enforce safe truncation on list fields to prevent validation errors and document visualizer in README

feat(cors): support wildcard * origin matching in development

build: track and commit config.yaml for deployment

build: include config.yaml in Docker build context

docs: add Hugging Face Space metadata to README.md

fix(gateway): implement self-healing JSON parser and repair engine for LLM outputs

Add deployment, CI, and final support regression coverage after the backend runtime is complete

Expose the recruiting agent through FastAPI with request tracing, readiness, CORS, and stable error responses

Add the stateless conversation-turn service that validates sessions and executes the orchestration graph

Add the LangGraph conversation workflow that connects analysis, retrieval, policy enforcement, revision, and fallback

Add the company configuration and initial outreach services before conversation-turn orchestration

Add deterministic conversation guardrails before wiring the orchestration graph

Add the OpenAI-compatible model gateway with workload routing and structured-output fallbacks

Add model-facing prompt builders that keep company context grounded, structured, and injection-resistant

Add deterministic in-memory retrieval for company evidence before introducing model-driven conversation behavior

Define the recruiting-agent domain contracts and validation rules before adding service behavior

Add the runtime configuration layer with safe YAML loading, dotenv precedence, and redacted settings diagnostics

chore: initialize backend repository tooling, packaging metadata, ignore rules, and local developer automation