Fullstack Developer with real LLM experience (CRM for a property developer)
Format: full-time · remote · product and team language — Russian
About the product
We are developing and enhancing a working CRM for the sales department of a residential real estate developer (Far East). Agents and managers use the system daily: they manage clients and deals, select apartments from live inventory, plan sales, and track lead sources. Integrations: amoCRM, Calltouch (call tracking), Telegram.
The task is not research demos, but AI functions that managers actually use every day.
Technology Stack
Frontend: Vue 3 (Options API), Quasar 2, Pinia, Vue Router, axios, ECharts
Backend: PHP (REST API)
Data/Integrations: amoCRM, Calltouch, Telegram bot AI: OpenAI / Anthropic API and/or open models (Llama, Qwen, GigaChat, YandexGPT)
Responsibilities
- Develop the CRM according to business requirements
- Design and implement LLM functions in the existing SPA and REST backend
- Build RAG over the internal "Knowledge Base" (assistant for sales scripts and agent questions)
- Integrate AI prompts into apartment selection (drag-and-drop proposal constructor based on budget/area/number of rooms/finishing)
- Perform auto-summarization of deals and clients, analysis of calls/leads from Calltouch
- Extract structured fields from raw manager notes
- Configure quality evaluation (evals), hallucination protection, secure UI output
Requirements (Mandatory)
- Real production experience with LLM: at least one custom AI feature brought to live users.
- Willingness to share what broke and how it was fixed (latency, cost, hallucinations, prompt regressions)
- RAG in practice: chunking, embeddings, vector storage (pgvector / Qdrant / Weaviate / Elastic), tuning output quality
- Function/tool calling and structured output: responses according to JSON schema, validation, retries for incorrect model responses
- Experience with providers: OpenAI and/or Anthropic; a plus — experience with open/on-prem models (Russian and sensitive data)
- Web stack for integration: solid experience with Vue/Quasar SPA and REST backend (PHP is a plus; if not, strong general backend experience and willingness to learn)
- Understanding of Russian language nuances in LLM: quality of prompts and eval sets in Russian, differences in tokenization/cost
- Evals and security: offline evaluation sets, prompt regression testing, working with PII, understanding prompt injection and mandatory sanitization of AI-generated HTML before rendering
Will be a plus
- Experience with amoCRM / Calltouch / Telegram Bot API
- Response streaming (SSE), caching (semantic/prompt caching), token cost consideration Speech-to-text (transcription and call analysis)
- Basic MLOps: prompt versioning, observability (Langfuse / Helicone / OpenTelemetry)
Conditions
Salary level — based on interview results, discussed individually
Format — remote, employment terms are negotiable
