ABOUT US Dragonfly is building the world's first Automated Solutions Architect. We help businesses navigate the complex landscape of modern tools (SaaS, AI, Infrastructure) by using an AI-powered platform that understands their unique context and recommends the optimal tech stack. Our platform is powered by a proprietary knowledge graph of 230K+ products, 4M+ companies, and the relationships between them — built from 100+ external data sources, LLM-driven research, and human curation. Every user-facing interaction — search, recommendations, diagram building, research — runs through AI systems that we design, build, and operate in-house. THE ROLE This is not a research role. This is a building role. You'll be writing the AI systems that power the product — recommendation engines, research pipelines, conversational agents, structured LLM orchestration — and shipping them to production. We don't separate "AI" from "engineering." The AI systems live in the same monorepo as the product, follow the same engineering practices, and ship through the same CI pipeline. You'll be working alongside product engineers in Python and TypeScript — the same codebase, the same review process, the same standards. If a product change is needed to ship your work — a new API endpoint, a streaming interface, a UI prototype to expose an agent — you make it. We've built the foundations: a recommendation engine with search, ranking, and requirement analysis; a conversational AI for interactive architecture diagrams; an autonomous research pipeline that enriches our knowledge graph; and structured output schemas for every AI interaction. We need someone to own these systems — improve what exists, build what's next, and push the boundaries of what's possible with applied AI. We operate a high-autonomy, high-trust environment. You'll be given a problem and the space to solve it — not a task list. We expect you to think beyond the immediate task — consider cost, latency, reliability, and how your work fits into the broader product. Curiosity matters: you should want to understand how everything connects, not just the model you're tuning. WHAT YOU'LL DO 1. RECOMMENDATION ENGINE The recommendation pipeline is the core of the product. Given a user's context, it retrieves candidate products, ranks them, analyses them against requirements, and streams results back in real time. You'll own the intelligence behind this. - Evolve how we rank and score products — the models that decide what gets recommended and why - Improve retrieval quality — query generation, embedding strategies, hybrid search - Experiment with new approaches to scoring, filtering, and personalisation - Optimise for cost and latency — these pipelines run on every user interaction 2. CONVERSATIONAL AI & AGENTS Architect is our interactive tool where users describe what they need in natural language and the platform researches, plans, and composes architecture diagrams in real time. Safety checks run concurrently. Results stream live. - Extend the conversational AI — new capabilities, better planning, richer context - Improve safety and guardrails systems - Build and ship new agent experiences as the product evolves - Design agent architectures that balance cost, quality, and latency 3. RESEARCH PIPELINE & KNOWLEDGE GRAPH Our research pipeline autonomously discovers, enriches, and classifies entities at scale — products, companies, and the relationships between them. It's how the knowledge graph grows and stays current. - Own the research pipeline end-to-end — improve coverage, accuracy, and throughput - Improve research quality — better prompts, validation loops, automated quality checks - Extend to new domains, new data sources, and new entity types - Build human-in-the-loop systems where AI proposes and humans approve 4. SEARCH, RETRIEVAL & EMBEDDINGS Every surface of the product depends on finding the right entities quickly. Search powers recommendations, diagram building, and the product catalogue. - Own and evolve our search and retrieval infrastructure — embedding models, hybrid search, ranking - Evaluate and integrate new embedding models as the field advances - Improve relevance across different search surfaces (recommendations, catalogue, conversational) - Web scraping and data extraction to enrich entities with live information 5. SELF-SERVING AGENTS & INTERNAL TOOLS We build agents that automate our own workflows — from orchestrating development tickets in parallel to maintaining the knowledge graph via chat commands with mandatory human approval. You'll build more of these. - Identify automation opportunities across the business and prototype solutions - Build internal agents that save the team hours of manual work - Ship proof-of-concept UIs when needed to expose agent capabilities to non-technical stakeholders 6. AI-NATIVE WORKFLOW & TEAM ENABLEMENT You are expected to keep up with the latest advancements in AI-powered development and proactively bring new tools, techniques, and workflows into the team. This isn't a nice-to-have — it's a core responsibility. - AI coding tools (Claude Code, Cursor, or similar) as your primary development environment - Trial new AI tools and techniques, and evangelise what works - Help the team adopt AI-native practices — pair with engineers, share workflows, raise the bar - Contribute to our AI-powered SDLC practices and tooling BOUNDARIES (SOFT, NOT HARD) Your primary focus is the AI systems, but you're not siloed. If shipping your work means writing a FastAPI endpoint, building a streaming interface, prototyping a React component to demo an agent, or writing a SQL transformation to feed a new feature — you do it. The codebase is a monorepo for a reason. You won't be the primary owner of the data platform, the frontend, or the infrastructure — but you'll touch all of them when the AI work requires it. TECH STACK - LL