Opis tej firmy nie jest jeszcze dostępny w tym języku.
Research Engineer, Agents
Vous concevrez des systèmes d’agents et des outils de recherche pour les équipes scientifiques de Boltz, puis les adapterez à de grands déploiements sur cluster. Le poste requiert une solide expérience en génie logiciel et dans la création de systèmes d’agents, ainsi qu’un jugement rigoureux sur la fiabilité de leurs résultats.
3× większe szanse na rozmowę kwalifikacyjną dzięki cold emailowi
Zobacz oryginalną ofertęInformacje o Boltz
- Branża
- Ochrona zdrowia, farmacja i biotechnologia
- Ostatnia runda finansowania
- Série A · 2026-11-01
- Ogłoszona kwota
- 110 000 000 $
- Aktualne oferty
- 12
Wymienieni inwestorzy
Szczegóły oferty
Pełny opis opublikowany przez Boltz.
About boltz
Boltz is a public benefit company building the next generation of AI-powered molecular modeling tools to make biology programmable and accelerate drug discovery, while keeping frontier capabilities broadly accessible. Boltz was founded by former MIT scientists who created Boltz-1, the first fully commercially available open-source model to approach AlphaFold 3-level accuracy in predicting the 3D structure of biomolecular complexes. We focus on two things: - Foundation models.
Our open models, Boltz-1, Boltz-2 and BoltzGen, are trusted by 100,000+ scientists across biotech and academia, and used in programs at every Top 20 pharma as well as leading agrichemical and industrial research organizations. Our latest proprietary foundation models power our products and the custom models we build with partners. - Boltz Lab and the Boltz API. Our products run our latest models, design pipelines (BoltzMol for small molecules, BoltzProt for protein design) and agents as reliable, production-grade tools built around real chemistry and biology workflows.
Teams start from a target and a hypothesis, then generate, evaluate, and rank candidate molecules in Boltz Lab, run our models from their own code and agents through the Boltz API, or deploy them inside their own environment. The two feed each other: our open releases earn scientific credibility and community insight, and our products turn our full range of models into real impact for partners including Pfizer, Takeda, GSK, Bristol Myers Squibb and DSM-Firmenich. You can read more about our mission, research and product vision in our manifesto https://boltz.bio/manifesto.
About the role As a Research Engineer (Agents), you will build and own the agent harnesses behind research at Boltz. You will design them for our scientific and research use cases, including agentic research pipelines and researcher tooling, and scale them to run on our cluster. You will be part of Research Engineering and work with researchers across our teams.
This role is ideal for a strong systems engineer who has built agent systems, has sharp intuitions for the capabilities and limitations of current frontier LLMs, both open- and closed-source, and is rigorous about the correctness of agent output. You do not need an ML research background.
What you’ll do
- Design and build agent harnesses for research: tool use, orchestration, and context and state management.
- Work with researchers to build agentic pipelines and tooling for their work, and make them reusable across teams.
- Scale these harnesses to support large agent swarms on our cluster.
- Build observability, verification and cost control for large agent runs.
- Evaluate agent setups rigorously, and keep our harnesses current as frontier LLMs evolve. About you
Minimum requirements
Strong software engineering experience, including distributed systems or infrastructure that runs many long-running jobs in parallel and handles failures.
- Experience building agent systems or harnesses (tool use, orchestration, context and state management, evaluation) that you can discuss in depth.
- Good judgment about the capabilities and failure modes of frontier LLMs, and about keeping agent-produced work correct and maintainable.
- A track record of building tools that were widely adopted by researchers or engineers.
Preferred requirements
Experience developing open-source agent harnesses (e.g. Pi, Codex, DSH) or extending existing ones (e.g. Claude Code).
- Understanding of LLM architectures, and experience fine-tuning LLMs with supervised fine-tuning (SFT) or reinforcement learning (e.g. RLHF, RLVR, GRPO).
- Background in NLP or information retrieval, such as search, ranking, embeddings, vector search or retrieval-augmented generation.
- Experience running workloads on a shared compute cluster or cloud infrastructure (e.g. Kubernetes).
- Exposure to scientific research workflows.
- Experience using Boltz models.
What we offer
- Work alongside one of the most talent-dense teams in the field.
- Significant ownership and independence, with responsibility for driving projects from concept to deployment.
- Highly competitive salary with meaningful equity in a public benefit company. -
Benefits
- 25 days plus bank holidays a year
- Annual team offsite trip
- Private healthcare (BUPA), covering pre-existing conditions
- Pension program
- Paid parental leave
- Wellhub gym membership
- Free lunch every day
- Travel expenses How we hire
- Initial screen with a member of the Talent team
- Diagnosis interview: work through a degraded training run with us (no manual coding test)
- Onsite in London or virtual onsite with the team
- Offer + Onboarding Our commitment Boltz is committed to building a diverse team and an inclusive workplace. We welcome applicants of all backgrounds. If you need any adjustments during the hiring process, tell us and we'll make it work.
Wymagania
- Solide expérience en génie logiciel, notamment en systèmes distribués ou en infrastructure exécutant de nombreux travaux
- Expérience approfondie dans la création de systèmes ou de cadres d’agents (utilisation d’outils, orchestration, gestion
- Bon jugement sur les capacités et les modes de défaillance des LLM de pointe, et sur la façon de maintenir la justesse e
- Expérience avérée dans la création d’outils largement adoptés par des chercheurs ou des ingénieurs
- Expérience dans le développement de cadres d’agents open source (p. ex. Pi, Codex, DSH) ou l’extension de cadres existan
- Compréhension des architectures LLM et expérience de leur réglage fin par apprentissage supervisé (SFT) ou apprentissage
- Expérience en TAL ou en recherche d’information, comme la recherche, le classement, les embeddings, la recherche vectori
- Expérience dans l’exécution de charges de travail sur un cluster de calcul partagé ou une infrastructure cloud (p. ex. K
- Exposition aux flux de travail de recherche scientifique
- Expérience dans l’utilisation des modèles Boltz
Wymienione benefity
- 25 jours plus jours fériés par an
- Voyage annuel de l’équipe
- Assurance santé privée (BUPA), couvrant les affections préexistantes
- Régime de retraite
- Congé parental rémunéré
- Abonnement à la salle Wellhub
- Déjeuner gratuit chaque jour
- Frais de déplacement
3× większe szanse na rozmowę kwalifikacyjną dzięki cold emailowi
Aplikuj na tę ofertę
Jak aplikować?
Oferta jest opublikowana na stronie Boltz. Otwórz oryginalną ofertę ↗ aby wysłać aplikację. Screeningpass nie wypełnia formularza za Ciebie.
Do kogo w Boltz napisać w sprawie tego stanowiska?
Rozpocznij wyszukiwanie kontaktu na tej stronie. Screeningpass zaproponuje Ci maksymalnie trzy osoby związane ze stanowiskiem. Wybierz osobę, do której chcesz napisać, a Screeningpass poszuka jej służbowego adresu e-mail i przygotuje wiadomość. Sprawdź ją i wyślij samodzielnie przez Gmail. Adres e-mail nie zawsze jest dostępny i nigdy nie ma gwarancji odpowiedzi.
Ile kosztuje wyszukanie kontaktu?
Jedno wyszukiwanie kosztuje 1 kredyt. Jeśli wyszukiwanie się nie powiedzie, kredyt wróci na Twoje konto. Za rejestrację otrzymasz 3 kredyty, czyli 3 wyszukiwania.
Do czego służy Screeningpass?
Screeningpass zbiera publikowane przez firmy oferty staży, kształcenia dualnego i pracy. Dla danej oferty Screeningpass znajduje osobę do kontaktu w firmie, szuka jej służbowego adresu e-mail i przygotowuje wiadomość na podstawie Twojego CV. Zobacz szczegółowo, jak to działa.