AI Engineer — LLM Features for Team Chat (Summarization, Smart Replies, AI Assistant)

We are building TeamChat, a workspace-based team collaboration platform (similar to Slack). This role owns the LLM-powered product features: thread summarization, smart replies, and an in-app AI assistant with tool use. We have a detailed scope document ready to share with shortlisted candidates. This is one of two AI roles we are hiring; strong performance leads to ongoing, long-term collaboration. CORE RESPONSIBILITIES & SCOPE OF WORK 1. Thread & Channel Summarization: On-demand and scheduled digests ("catch me up") with map-reduce summarization for long threads, multilingual content handling (EN/JA), and cost-aware model routing (cheap model first, escalate when needed). 2. Smart Replies & Drafting: Context-aware reply suggestions and message drafting that respect channel tone; perceived latency under ~2s via streaming. 3. In-App AI Assistant: Conversational assistant with tool use / function calling (search workspace via the RAG service, summarize, draft, set reminders), human-approval steps for sensitive actions, structured outputs. 4. LLM Orchestration Layer: Provider-agnostic abstraction over Anthropic Claude / OpenAI APIs with retries, fallbacks, token budgeting, per-workspace rate limits, and full cost tracking per feature. 5. Evaluation & Safety: Offline eval sets for summarization quality, prompt-injection hardening for user-generated content, PII-aware logging. 6. Delivery: Python service with documented internal APIs; tests + eval harness included. REQUIRED TECH STACK - Python 3.11+, FastAPI - LLM APIs: Anthropic Claude, OpenAI (function calling / tool use, streaming, structured outputs) - Orchestration: thin hand-rolled layer preferred (LangChain experience fine) - Redis/Celery workers, PostgreSQL PROJECT DETAILS - Engagement: Hourly, $15–$25/hr depending on experience. ~30 hrs/week, initial 3 months, ongoing long-term for the right person. - Process: Daily async standup (English, text), code review via GitHub PRs, 2-week sprints. At least 3–4 hours of overlap with JST (UTC+9). - IP & Code: All code delivered in our GitHub org from day one; full source ownership by us. - Language: English required. Urdu-speaking developers welcome. WHO SHOULD APPLY Please do NOT apply if your experience is limited to basic chatbot demos, simple OpenAI API wrappers, or tutorial-level LangChain projects. We will ask about production metrics (cost, latency, output quality). QUESTIONS TO ANSWER IN YOUR PROPOSAL 1. Describe an LLM-powered product feature you shipped to real users. What did it cost per user per month, and how did you reduce that cost? 2. Describe an agent/tool-use implementation you built: which tools, how you validated tool inputs, and how you handled failures. 3. What is your approach to evaluating summarization quality beyond "it looks good"? 4. GitHub/portfolio links, timezone, weekly availability, proposed rate. 5. Start your proposal with the word TEAMCHAT.