Software Engineering Intern - Product AI (Winter 2027)

BitGo is the leading infrastructure provider of digital asset solutions, delivering custody, wallets, staking, trading, financing, and settlement services from regulated cold storage. Since our founding in 2013, we have focused on enabling our clients to securely navigate the digital asset space. With a global presence and multiple Trust companies, BitGo serves thousands of institutions, including many of the industry's top brands, exchanges, and platforms, and millions of retail investors worldwide. As the operational backbone of the digital economy, BitGo handles a significant portion of Bitcoin network transactions and is the largest independent digital asset custodian, and staking provider, in the world. For more information, visit www.bitgo.com.

Product AI exists to make BitGo easier to adopt and more operationally valuable to use. We believe the fastest path there is letting customers ask for what they want in natural language and, over time, letting the product carry out that work for them under confirmation and audit. We treat that as a hypothesis to prove with experiments and measurable customer outcomes.

We are the customer-facing half of BitGo’s AI work. AI Platform owns the base infrastructure, including model access, hosting, and shared runtime. Product AI owns what customers touch: the product surfaces, workflows, context that makes answers correct, and the evaluation and governance applied to those use cases. If you want to run inference infrastructure, that is the other team. If you want to own whether a customer got the right answer and accomplished the right thing, this is the one.

Right now that means taking our AI Assistant – BitGo Assist – beyond question-answer and navigation towards a full autonomous assistant, capable of acting arbitrarily on your behalf. What comes after—multi-step actions with confirmation and audit, proactive triggers, and context-aware workflows across surfaces such as trading, borrowing, settlement, and wallets—is sequenced by what each experiment teaches us. The team’s north star is improving customer activation, engagement, and retention by helping customers accomplish more of their operational work inside BitGo.

We work in two-week cycles against a roadmap you can actually read, shipping small vertical slices in front of customers. We watch leading indicators weekly and measure both the outcomes we deliver and the AI cost of delivering them.

Responsibilities

• Turn AI product hypotheses into experiments that settle them. Pick high-value use cases, ship the smallest version that produces a real signal, and report what it proved—including when the idea does not work.
• Build and operate customer-facing AI surfaces and workflows across the BitGo application (trade, policies, custody, etc), support experiences, Developer Portal, and related product areas.
• Evolve BitGo Assist from reliable read-only answers toward multi-step workflows that can take action with explicit confirmation, auditability, and appropriate proactive automation.
• Own correctness where it counts. When a customer asks about a balance, transaction state, or API response, the answer is either right or it is a trust problem. Build the evaluation sets, regression tests, and prompt and model tuning loops that keep answers grounded as models and prompts change.
• Close the loop with Support and internal tooling teams: measure whether an answer was correct, whether the customer was routed to the right place among Assist, Scout, and a human, and whether the question was genuinely resolved rather than merely deflected. Count autonomous resolution and time and cost saved; do not count a frustrated customer who gave up.
• Instrument every AI surface so its cost, quality, and latency are continuously visible in MLflow. Drive latency down by an order of magnitude without trading away answer quality.
• Build reusable safety controls: prompt-injection defense, PII and data redaction, tenant isolation, and policy enforcement applied consistently across product surfaces so teams do not have to reinvent them.
• Consolidate duplicated context repositories and knowledge bases into reusable BitGo context, integrating API specifications, documentation, product data, and user and account context safely.
• Manage production concerns including model and provider selection, token usage, quotas, noisy neighbors, latency, and graceful degradation.
• Partner with Product leadership and PMs across the company to find AI opportunities in their areas, provide the frameworks and tools to evaluate them, and personally build the highest-value use cases.
• Work with AI Platform at the infrastructure boundary while owning how shared capabilities land in customer-facing product experiences.
• Communicate progress, tradeoffs, experiments, and outcomes clearly to technical and non-technical partners.

Required

• Experience building and operating production backend or full-stack systems, including APIs and distributed or multi-step workflows.
• Product sense: the ability to turn a vague opportunity into a hypothesis, a small vertical slice, and a measurement that can settle it.
• Ability to reason about correctness, nondeterminism, latency, cost, security, and failure modes in customer-facing AI systems.
• Experience with observability and operational practices for production systems.
• Practical understanding of AI security and governance concerns, including prompt injection, sensitive-data exposure, tenant isolation, and policy enforcement.
• Ability to drive projects end to end, from discovery and design through implementation, rollout, instrumentation, and iteration.
• Willingness to treat security and observability as part of shipping, not a later phase.
• Ability to work directly with customers, Product, Support, and engineering partners.
• Excellent written communication: your reasoning should remain clear even when you are not in the room.

Preferred

• Experience building customer-facing AI assistants, search experiences, support workflows, or agentic product experiences.
• Experience designing agentic systems that take multi-step action under human confirmation and audit.
• Experience improving support resolution through accurate answers, appropriate escalation, and autonomous handling of requests.
• Experience with MLflow or comparable tracing and evaluation tooling for cost, quality, and latency.
• Experience operating AI products where provider or model routing, token usage, quotas, latency, and cost must be balanced against quality.
• Familiarity with MCP, CLI, SDK-based access surfaces, or other agent integration patterns.
• Experience consolidating fragmented knowledge bases or retrieval context into a reusable layer.
• Background in fintech, crypto, custody, or another regulated domain where a wrong answer has serious consequences.
• Familiarity with financial product domains such as trading, borrowing, settlement, wallets, onboarding, or developer platforms.
• Experience working in fast customer-feedback cycles and turning early internal or external usage into reusable product patterns.

If you will be working from Canada: the current target rate for 2027 interns in Canada is $60 CAD per hour. If the market indicates this needs to change, we would share this with you during your interview process.

*Please note that we are required by law to pay you in the currency of the country you will be working from.*