Senior AI Engineer / Forward Deployed Engineer

Towards AI, Inc.; Greater London; Hybrid; GBP 80,000 - 100,000; Full time

Job summary

Towards AI, Inc. is seeking a highly skilled engineer to lead the technical delivery of custom AI systems for our high-profile clients. You'll be responsible for end-to-end architecture, build, and deployment of LLM applications, working closely with client teams.

The ideal candidate has extensive experience in AI engineering, strong proficiency in Python, and a dedication to quality through mentorship and rigorous standards. Join us to shape the future of AI in finance.

Qualifications

• 5+ years professional software engineering experience.
• 2+ years building LLM applications, RAG systems, or AI workflow automation.
• Daily use of agentic coding tools like Claude Code.

Responsibilities

• Own technical delivery of custom AI systems end-to-end.
• Build production LLM applications using various APIs.
• Mentor junior engineers and maintain technical quality.

Skills

Python

LLM applications

RAG systems

Docker

Git

AI experience in finance

Technical communication

Tools

OpenAI

Claude Code

Gemini

React

Job description

We are building a premium AI consultancy targeting an underserved whitespace: transforming small-to-mid-cap portfolio companies and finance firms that cannot afford to get AI wrong. Around 95% of enterprise AI experiments never reach production. We hire only top-tier talent to close that gap for clients with serious engagement budgets and real urgency to ship.

Six years in: 500,000 learners since 2019, an O'Reilly book referenced inside Fortune 500s, and a 15-person engineering team shipping live work for clients including J.P. Morgan, Intel, Europol, NYPL, Nviya-Prime, Maoki, and Activeloop.

You will own the technical delivery of custom AI systems end-to-end: architecture, build, evaluation, hardening, and production rollout. Typical builds include agentic systems for investment research and due diligence, retrieval-augmented copilots for finance workflows, document analysis pipelines, market mapping tools, internal knowledge assistants, systems integrations, and workflow automations for portfolio operations. Most weeks are heads-down engineering with your pod, shipping systems that real client teams depend on day-to-day.

Nviya-Prime

We architected the Agentic Intelligence Engine for Nviya-Prime, a B2B FinTech replicating institutional-grade banker reasoning. Three coordinated sub-agent services orchestrate ~800 prompts per client analysis. Event Targeting classifies material signals across economic, geopolitical, regulatory, market, supply-chain, and idiosyncratic domains. Predictive Scanning runs 24/7, scoring live events on materiality x urgency and surfacing prioritised next-best-actions, for example an FX hedge plus interest-rate contract in response to a funding-cost spike.

Production stack: agentic retrieval-augmented generation (RAG), hybrid search, contextual retrieval, reranking, model routing, LLM-as-judge evaluation, end-to-end observability. Currently shipping into live demos with global banks across Europe.

Two modes, depending on the project.

For most engagements you work from our team alongside your engineering pod, with regular technical touchpoints into the client's engineering and data teams. A deployment strategist on each engagement owns the executive client relationship and the bulk of stakeholder management, leaving you to focus on getting the system built right.

For our largest multi-quarter engagements, typically with financial services firms, PE houses, or strategic portfolio companies, you go forward-deployed at the client site for kickoff sprints, key build phases, and production rollout. Your direct client counterparts are engineering, data, product, and end-user teams. You should be excited by this mode of working: it offers the fastest feedback loops and the most direct influence on what gets shipped of any client setup we run.

• - Lead technical scoping and architecture across client engagements.
• - Build production LLM applications using OpenAI, Anthropic, Gemini, Bedrock, Vertex AI, or similar.
• - Architect RAG systems: retrieval quality, grounding, source attribution, metadata, hybrid search, reranking, access control.
• - Build agentic workflows: tool use, structured outputs, orchestration, retries, fallbacks, state, human review.
• - Design eval frameworks for prompts, retrieval, agents, models, regressions, safety, latency, and cost.
• - Define guardrails for privacy, governance, prompt injection, sensitive tool use, and hallucination risk.
• - Deploy and monitor with Docker, cloud infrastructure, continuous integration and deployment (CI/CD), logging, and observability.
• - Pair with client engineering and data teams on integrations, deployment, and technical handover.
• - Mentor junior engineers; hold the technical quality bar across delivery.
• - Ship reusable firm assets: reference architectures, prompt libraries, eval harnesses, agent skills, implementation playbooks.

Daily, expert use of Claude Code, Codex, Cursor, or similar agentic tools is non-negotiable. Planning, codebase exploration, implementation, testing, refactoring, documentation, research, debugging: all of it. Engineers who already work this way outpace traditional engineers by a wide margin, and that gap is the bar for this role.

The standard is intelligent supervision, not blind delegation. You give the agent the right context, constrain the task, inspect the diff, run the tests, catch weak assumptions, and decide when to take the keyboard back. You ship at speed but never ship what you have not understood.

You should also actively design reusable AI infrastructure for the firm: Claude Skills, Model Context Protocol (MCP) servers, sub-agents, eval harnesses, prompt patterns, and workflow templates. Our strongest engineers build the tooling that makes the rest of the team faster.

• - 5+ years professional software engineering experience.
• - 2+ years building LLM applications, RAG systems, AI agents, or AI workflow automation in production.
• - Shipped at least two AI systems beyond simple chat: domain-specific RAG, multi-tool agents, document analysis, internal copilots, or comparable.
• - Strong Python, plus production experience with application programming interfaces (APIs), integrations, and data-heavy systems.
• - Hands-on with OpenAI, Anthropic, Gemini, Bedrock, Vertex AI, or comparable LLM APIs.
• - Real RAG depth: embeddings, chunking, indexing, metadata, hybrid search, reranking, retrieval evaluation, grounding.
• - Real agent depth: tool calling, structured outputs, orchestration, state, retries, safe execution boundaries.
• - Strong eval discipline: golden sets, regression tests, human review loops, considered use of LLM-as-judge, release gates, production monitoring.
• - Production judgment on latency, cost, scalability, reliability, privacy, security, and data governance.
• - Comfort with Docker, Git, CI/CD, cloud deployment, databases, logging, monitoring.
• - Daily, expert use of agentic coding tools (Claude Code, Codex, Cursor) for real engineering, with careful inspection of what they produce.
• - Strong technical communication. You can explain architecture, trade-offs, and risk to engineering counterparts and to your own pod, and hold your own in a senior client conversation when the strategist hands you the floor.
• - Prior work with investors, PE firms, venture funds, family offices, or portfolio companies.
• - AI experience in investment research, due diligence, finance, sales, customer support, reporting, or knowledge management.
• - TypeScript, React, Next.js, Postgres, pgvector, Pinecone, Weaviate, or Qdrant.
• - LangChain, LangGraph, LlamaIndex, LangSmith, Langfuse, Braintrust, OpenTelemetry, MCP servers, OpenAI Agents SDK, or Claude Skills.
• - Solutions engineering, customer engineering, founder-led product, or enterprise implementation backgrounds.
• - Fine-tuning, distillation, or open-weight model experience.

A builder and a field engineer. You take an ambiguous technical problem, identify the constraint that matters, and design a system that gets used rather than admired. You move between architecture and code without losing either. You know when an agent is the right shape, when deterministic code is better, when retrieval is the bottleneck, and when a proposed scope is the wrong one. You are AI-native enough that work without these tools feels broken.

You should be able to describe a system you have built in concrete terms: problem, architecture, model choices, retrieval design, evals, failure modes, cost profile, rollout plan, and what you would change next time.

Towards AI is an AI education and development company founded in 2019, with 500,000 learners reached, 120k newsletter subscribers, 80k Discord members, and 8,000+ copies sold of our O'Reilly book Building LLMs for Production. We now operate a specialist 15-person AI engineering team focused on investment firms, PE, portfolio companies, and finance firms in regulated industries.

Co-founded by Louie Peters (ex-J.P. Morgan VP, credit research) and Louis-Francois Bouchard (ex-Mila, Polytechnique Montreal). Our deep roots in AI research and education mean our engineers stay at the forefront of the field, with direct exposure to the latest academic developments and the unique opportunity to apply cutting-edge AI in real-world, high-stakes environments from day one.