← All vacancies

Middle AI Software Engineer

Infrastructure · Work from the office · Relocation paid by the company

Full time · 3+ years of experience

We are looking for a Middle AI Software Engineer (Python / C++): a person who will go with us from the point "an engineer increases the business result while spending a few thousand dollars a day on tokens" to "multiplies the business result of the whole team, even when spending 10k+ dollars a day on tokens".

You gave an agent a task, and in twenty minutes it produced "working" code that builds and passes the tests. But hidden in it are a subtle bug, an invented API and a couple of suboptimal decisions. Noticing this quickly, understanding why the agent did it and bringing the code into shape is the skill we are looking for. The agent speeds the work up many times, but the responsibility for every line stays with the engineer.

Can you do that? Then we have something to talk about.

We build trading infrastructure for high-frequency trading (HFT): a large system that has to run stably with minimal latency at critical moments, under 10x the usual load. For this we use our own user-space network stack, sub-microsecond IPC, optimised data formats and more.

What you will do

A Middle AI Software Engineer here is a strong Python and C++ engineer who has mastered agent tools and uses them both to do more on their own and to raise the productivity of the whole team. This is not a "prompt engineer": it is an SWE whose AI skills complement a solid engineering base rather than replace it.

Example tasks

  • Integrating LLMs into services where it is justified. Tool use / structured output through the API, with control over execution; knowing which tasks to give to a model and which not
  • Full HFT tasks with agents as a multiplier. Connecting new exchanges, engineering research of exchange infrastructure, latency optimisation
  • Developing internal agent tooling. Agent rules and instructions for our codebase, custom tools for agents, MCP servers, evals: the infrastructure that makes the team's use of agents more effective
  • Building agent workflows. Assembling pipelines where agents decompose a task, work through tasks and verify each other's result. Choosing what to delegate and how, and bringing these processes to a state where the whole team uses them

What matters for this role

  • Use agent coding tools (Claude Code, Codex, Cursor) daily on real tasks and have already achieved commercial results
  • See where an agent speeds things up and where it adds hidden tech debt, and why an output is efficient or not, at the level of caches, pipeline, memory model
  • Quickly find subtle bugs, hallucinations and problems in generated code, and take full responsibility for the result, as for code written by hand
  • Understand LLMs at an engineering level: context, tokens, the causes of hallucinations, tool use, how to measure and control output quality
  • Are strong in Python and can work confidently in a C++ codebase
  • Understand how the OS and hardware work (CPU, RAM, NIC), network protocols (Ethernet, UDP, TCP) and the basics of multithreaded programming
  • Real skills matter more to us than a formal number of years in the role, but these tasks usually suit engineers with 3+ years of commercial server-side development, including production experience in C++.

Nice to have

  • Building agent pipelines or multi-agent orchestration, MCP servers, evals
  • Integrating LLMs into your own services through the API: tool use / function calling, structured output, streaming, cost control
  • Setting up AI tools for a large codebase: rules, custom tools/hooks, indexing
  • HFT, order routing, market data processing, exchange research
  • An environment for deep work. No meetings for the sake of meetings, short feedback loops, performance coaching for personal productivity
  • Profit sharing. Your income will reflect your personal contribution to the team's result (avg +50−100% of annual income)
  • A strong team. Engineers from tier-1 companies and experienced industry specialists, winners and prize winners of olympiads in mathematics, programming and physics at the level of IMO, IOI, the All-Russian Olympiad and the ICPC semifinal
  • Full care for employees: ー monitoring and help with keeping key health indicators ー health insurance with dental care ー company events in different parts of the world ー monthly events: sports and wellness activities from the company ー food in the office from the best restaurants
  • Tasks that directly affect PnL. We have no "not very important" tasks: we organise the work so that your efficiency → 1
  • An unlimited token budget. If a quality result takes spending $10k in a day, do it
  • Personal responsibility for the result. Your decisions go to production and affect PnL. With every project the area of responsibility grows, and with it the freedom to choose the approach

Grow with the scale of the tasks

At Spectral there are no rigid boundaries by title: you can propose projects and widen your area of responsibility. For growth there is a personal track, regular reviews and the support of an experienced team.

  • A strong engineering environment

    Next to you are industry specialists, engineers from tier-1 companies and olympiad winners. We share knowledge and work towards a common goal.

  • Resources for an idea

    Hardware, training and computing power for the tasks. The chance to start a project and own its result.

  • Productivity support

    Performance coaching: goals, motivation and handling the workload. We help you keep focus and recover.

Conditions that help you focus

  • Income

    Salary, profit sharing without a ceiling for technical roles, bonuses and awards.

  • Relocation

    Tickets, the first month of housing, documents and the move of your family.

  • Health and life

    Health insurance with dental care, sport, meals in the office and team trips.

The full spectrum to win

  • Income without limits

    Salary, bonuses and awards, profit sharing from the common PnL without a ceiling for technical roles. Responsibility grows with every project, and income grows with it.

  • Personal development track

    Performance coaching: goals, motivation and managing workload, and with them sleep, sport and nutrition.

    Regular reviews and mentors from the industry. Equipment, education, languages and any computing power the work needs.

  • Health, wellness, relocation

    Health insurance with dental care, discounts on insurance for family, medical tests every six months. Breakfasts and lunches in the office, gyms and courts. We hire from any country: tickets, housing for the first month, documents and moving the family

  • Beyond the office

    Friday dinners, monthly events and internal competitions. Every six months the whole team travels to a new country. Private events attended by celebrities, and other exclusive events

How hiring works: role, track and timeline

  • A role for your strengths

    We look at your experience, strengths and interests and pick a role to match them. We put the interview track together the same way, so at the interviews you talk about what you are strong in.

  • Accelerated track

    The track usually takes about three weeks. If timing matters, tell the recruiter and we will run all the interviews within a few days

  • Only what the work needs

    We test the knowledge and the approach to problems that the role needs. Each section is tied to what you will be doing.

From first meeting to offer

  1. 01HR interview
  2. 022–3 technical interviews
  3. 03Fit interview
  4. 04HR call
  5. 05On-site interview with the team and the founder
  6. 06Offer

The number of stages depends on the role and your experience — we agree the track at the first conversation

Still have questions?

    Let's discuss the next steps

    Prefer a personal conversation?

    Write directly

    Middle AI Software Engineer

    Fields marked with an asterisk are required