AI INTEGRATOR / LLM APPLICATION DEVELOPER / 003
AI-NATIVE
DEVELOPMENT.
3+ YEARS / GPT-BASED DELIVERY
3+ years of continuous GPT-based software delivery. I use AI as an implementation layer inside a structured development process: requirements → implementation → manual validation → regression analysis → corrective specification → new iteration → deployment → release acceptance.
I have been working with GPT-based development workflows since 2023, progressively integrating new generations of AI systems into increasingly complex product delivery. My scope now extends beyond AI-assisted coding into hands-on LLM integration, local model runtime configuration, retrieval/FAQ systems, multilingual assistant behavior, data and feedback layers, external API tooling and operational support automation.
Hands-on AI Integrator / LLM Application Developer scope beyond AI-assisted coding.
Hands-on launch, runtime configuration and practical tuning of local LLM environments including Qwen-based models.
Built and tuned project-documentation knowledge flows with configurable retrieval, context and generation controls rather than relying on a single static prompt.
Designed assistant character/persona behavior: system instructions, tone, response boundaries, context rules and consistent virtual-character behavior.
Integrated LLM capabilities into practical product and support workflows, then validated outputs through hands-on QA, iteration and release acceptance.
Concrete engineering evidence from the reviewed AI assistant subsystem snapshot.
Four structured support flows — ban appeals, technical issues, player reports and purchases — with intent detection, field extraction, multi-turn state and ticket creation through the existing Discord support module.
Russian, Ukrainian and English detection plus explicit output-language enforcement and corrective regeneration when model output violates the selected language.
SQLite for per-server configuration/static FAQ plus MySQL for dynamic FAQ and answer feedback, with automatic schema/index creation, usage counters, reaction-based promotion and manual-review escalation flags.
Configurable retrieval thresholds, top-K/context limits, knowledge-text chunking, source tags and groundedness scoring that validates referenced source IDs and lexical overlap with retrieved context.
Steam profile lookup, SteamID64 and vanity-URL resolution, profile confirmation and structured identity data integrated into support/report routing before ticket creation.
Per-server channel, trainer role, system prompt and MySQL configuration; static FAQ CRUD/reset; database credential testing, pooled connections, cache invalidation and runtime status visibility.
Product definition, controlled AI implementation and release quality remain under direct personal ownership.
Business logic, workflow, priority and expected outcome.
Detailed implementation briefs, edge cases, constraints and acceptance criteria.
Context preparation, implementation tasks, patch iteration, debugging and controlled refinement.
Information hierarchy, workflows, interface behavior and acceptance of visual result.
Personal hands-on testing of every significant feature and patch.
Verification that new changes preserve existing behavior and that model responses respect language, grounding, policy and workflow constraints.
Installation, environment checks, rollback readiness and final acceptance.
AI output is never treated as automatically correct. Before delivery I validate functional behavior, edge cases, regressions, acceptance criteria, language/grounding constraints, integration behavior and deployment / rollback readiness.