Qureos Mentor

Samurai Ops

About the Role We build production AI agents and LLM-powered systems not demos or prototypes that stop at a notebook, but tools that run reliably, handle real user data, and get used every day. Recent builds include a multi-node LangGraph pipeline that personalizes cold outreach end-to-end, a RAG-backed comment generator that grounds output in thousands of real examples plus live web research, an AI career-coaching copilot with resume parsing and job-matching agents, and a reply-handling engine that classifies conversation state and drafts on-brand responses across multiple personas. We're looking for an engineer who can own this kind of system from architecture to deployment.What You'll Work On- Agentic pipelines: Multi-step LangGraph agents designing stateful, multi-node pipelines with conditional routing, revision/quality loops, and tool calling (not single-shot prompt calls). - Backend services: FastAPI services that expose these agents as real APIs, request/response schemas, async endpoints, background jobs, and integration with Streamlit or other front-ends. - RAG systems: Building retrieval pipelines (ChromaDB, Pinecone, or similar) over domain-specific data embedding strategy, chunking, and retrieval tuning that actually improves output quality, not just &quot;add a vector store.&quot; - LLM workflow &amp; prompt engineering: Going beyond &quot;write a good prompt&quot; building structured, multi-pass prompting strategies, scoring/revision loops, and humanization layers that make LLM output pass as natural rather than generic. - Data enrichment &amp; integrations: Pulling in external signals (LinkedIn data, web search APIs, job boards) and turning them into structured inputs the agents can reason over. - Data modeling: Designing schemas (SQL + vector) for user profiles, analysis caching, and conversation/session state so we're not re-computing expensive LLM calls. - Observability: Using LangSmith (or equivalent) to trace, debug, and evaluate agent runs this isn't optional once a pipeline has more than a couple of nodes.Skills We Need Core (must-have)Python: Strong, production-level Python. LangGraph / LangChain: Hands-on experience building actual graphs multiple nodes, conditional edges, state schemas not just LangChain chains. FastAPI: Comfortable designing and shipping REST APIs, async patterns, Pydantic models, dependency injection. RAG / vector search: Practical experience with at least one vector store (ChromaDB, Pinecone, Qdrant, Weaviate) and building retrieval that's actually evaluated, not just wired up. LLM APIs: Comfortable working directly with the Anthropic and/or OpenAI APIs tool use, structured outputs, system prompt design, model selection by task. Relational databases: SQL schema design plus comfort with at least one ORM (SQLAlchemy or similar). Systems thinking: Can read a vague product idea (&quot;analyze the post, then generate 3 comment variations&quot;) and turn it into an architecture diagram and a state machine.Strong Plus LangSmith or similar tracing/eval tooling. Experience with output &quot;humanization&quot; making generated text pass as natural, on-voice, and non-repetitive across many samples. Docker + basic deployment (Compose, CI/CD) for shipping agents as services. Familiarity with Streamlit or a lightweight frontend for internal tools/demos. Experience with web scraping/enrichment APIs (LinkedIn data providers, Google Custom Search, job board APIs). Redis or similar for caching expensive LLM calls and managing session state.What Good Looks Like Here You think in pipelines and state machines, not single prompts you instinctively break a fuzzy task into nodes, routing logic, and failure/retry paths. You care about whether the output is actually good you'll build a scoring or revision loop rather than ship the first draft an LLM produces. You design for reuse caching analysis results, avoiding redundant LLM calls, and structuring state so the next feature doesn't require a rewrite. You're comfortable being handed a one-paragraph idea and coming back with a real architecture, not just clarifying questions.Tech Stack Snapshot Languages Python (primary), JavaScript/TypeScript (frontend touchpoints) Agent frameworks LangGraph, LangChain Backend FastAPI, Pydantic, async/await Vector / retrieval ChromaDB (also open to Pinecone, Qdrant) Data PostgreSQL / SQL, Redis (caching), SQLAlchemy LLM providers Anthropic Claude (primary), OpenAI Observability LangSmith Frontend / internal tools Streamlit, Next.js (for client-facing products) Infra Docker, Docker ComposeExample Projects You'd Be Building On / Alongside Outreach personalizer: A cold-outreach personalization agent 16-node LangGraph pipeline that researches a prospect, drafts a message, scores it against a quality bar, and revises until it clears threshold. Comment / content generator: A RAG-grounded generation tool that classifies a target post, retrieves from a bank of ~1,600 historical examples, optionally pulls live web research, and produces multiple distinct angles, then humanizes the output. Coaching platform: Resume parsing, career-match, job-match, and conversational copilot agents sharing a common tool registry and data layer. Reply handling engine: A conversation-state classifier that reads an inbound message thread, identifies which of ~10 prospect states it's in, and drafts three meaningfully different reply options with persona-specific voice and case studies.Pay: ₹300,000.00 - ₹400,000.00 per yearWork Location: In person

AI Applications Engineer

Join Samurai Ops and explore 1   exciting career opportunities. They're hiring for various roles with options for remote or on-site work. Apply today and become part of their innovative team.

Samurai Ops

About Company

AI Applications Engineer