AI / GenAI Solutions Engineer

Soum is building an AI native layer across our C2C marketplace, from customer conversations to automated order handling, personalised discovery, recommendations, fraud signals, and seller tooling.

We're looking for a Senior GenAI Engineer to design and ship production-grade AI systems that touch millions of users across the buying and selling journey.

You'll work end-to-end: architecting LLM-powered services on our Python backend, integrating them into product surfaces on our React frontend, and partnering with teams across the company to find where AI genuinely moves the needle.

This is a senior builder's role with high autonomy, broad scope, and real impact on the business.

Soum is building an AI native layer across our C2C marketplace, from customer conversations to automated order handling, personalised discovery, recommendations, fraud signals, and seller tooling.

We're looking for a Senior GenAI Engineer to design and ship production-grade AI systems that touch millions of users across the buying and selling journey.

You'll work end-to-end: architecting LLM-powered services on our Python backend, integrating them into product surfaces on our React frontend, and partnering with teams across the company to find where AI genuinely moves the needle.

This is a senior builder's role with high autonomy, broad scope, and real impact on the business.

• Architect production GenAI systems across multiple domains, including conversational agents, automated order and dispute workflows, personalized discovery and recommendations, content generation, search relevance, and emerging use cases.
• Own features end-to-end , from problem framing and model selection through backend services (FastAPI, Python), frontend integration (React, TypeScript), evaluation, deployment, and monitoring.
• Design agentic workflows with tool calling, multi-step reasoning, retrieval augmented generation, and integrations with internal APIs, third-party SaaS, and event-driven systems.
• Build the retrieval and embeddings stack , including chunking strategies, embedding model selection, vector indexes, hybrid search, reranking, and retrieval evaluation pipelines.
• Make it reliable and cost-efficient through streaming, prompt caching, latency budgets, token cost optimization, observability for LLM calls, and graceful fallback when models or upstreams misbehave.
• Establish evaluation rigor with offline and online evals covering response quality, tool call correctness, hallucination rate, retrieval precision, and business KPIs.
• Drive experimentation and research by evaluating new models, frameworks, and agent patterns, running focused experiments, and bringing what works into production.
• Mentor and raise the bar for engineers across the team on AI and ML best practices, prompt engineering, and production readiness.
• Partner cross-functionally with Product, Engineering, Data, Ops, and CX to identify high-leverage AI opportunities and ship them.

A non-exhaustive list of areas we are actively building in or want to:

• Conversational AI for customer support, dispute resolution, and seller assistance

Conversational AI for customer support, dispute resolution, and seller assistance

• Automated order handling, escalation routing, and workflow orchestration

Automated order handling, escalation routing, and workflow orchestration

• Personalized discovery, recommendations, and search relevance

Personalized discovery, recommendations, and search relevance

• Listing quality, including auto-generated titles, descriptions, categorization, and image understanding

Listing quality, including auto-generated titles, descriptions, categorization, and image understanding

• Trust and safety, including fraud signals, anomaly detection, and content moderation

Trust and safety, including fraud signals, anomaly detection, and content moderation

• Internal agent tooling for ops and CX teams

Internal agent tooling for ops and CX teams

• 5+ years building production software , with at least 2 years shipping LLM and ML-powered features at scale.
• Strong Python for backend services, scripting, data pipelines, and ML tooling.
• Working ability in TypeScript and React for integrating AI features directly into product surfaces.
• Production experience with major LLM providers (Gemini, Claude, GPT) covering tool and function calling, structured outputs, streaming, prompt caching, and cost control.
• Deep understanding of Retrieval Augmented Generation (RAG) : document ingestion, chunking, embedding generation, vector databases (pgvector, Pinecone, Weaviate, or similar), hybrid retrieval, and reranking.
• Solid grounding in embeddings and vector search : dense vs. sparse representations, similarity metrics, indexing strategies (HNSW, IVF), and dimensionality tradeoffs.
• Strong ML and NLP fundamentals : transformer architectures, tokenization, fine-tuning vs. prompting tradeoffs, classification, ranking, and evaluation methodology.
• Experience with scripting and automation for data preparation, model evaluation harnesses, and offline analysis.
• Comfortable with relational databases (PostgreSQL), caching layers (Redis), REST, SSE, WebSockets, and event-driven architectures.
• Production experience with observability, A/B testing, and rolling out model changes safely.
• Experience with recommendation systems, learning to rank, or search relevance at scale.
• Marketplace or C2C background covering buyer and seller dynamics, disputes, fraud, and payouts.
• Multimodal model experience, including vision and image understanding for listings.
• Arabic NLP or bilingual product experience.
• Experience with agent frameworks (LangGraph, custom orchestrators) and the judgment to know when to use them.
• Fine-tuning, LoRA, distillation, or hosting open weight models in production.
• Open source contributions to LLM tooling, eval frameworks, or retrieval libraries.
• Ship over the architect. Lean code, no premature abstractions, no half-finished frameworks.
• Measurement-driven development. Features ship with evals and metrics, not vibes.
• Ownership. From idea to deployment to monitoring the first real users.
• Curiosity and range. This role spans many problem domains, and we want someone energized by that.