Asks for

The ad says“Strong fundamentals in RAG architectures, prompt engineering, agentic frameworks, and LLM orchestration (e.g., multi-step chains, tool use, structured output extraction).”
The ad says“Daily, hands-on use of modern coding harnesses (Claude Code, Codex, Cursor, or equivalent) — not as autocomplete, but as a primary way of shipping production code.”
The ad says“Daily, hands-on use of modern coding harnesses (Claude Code, Codex, Cursor, or equivalent) — not as autocomplete, but as a primary way of shipping production code.”
The ad says“Proficient in Python, FastAPI, PostgreSQL, containerization and scaling microservices.”
The ad says“Daily, hands-on use of modern coding harnesses (Claude Code, Codex, Cursor, or equivalent) — not as autocomplete, but as a primary way of shipping production code.”
The ad says“Proficient in Python, FastAPI, PostgreSQL, containerization and scaling microservices.”
The ad says“Strong fundamentals in RAG architectures, prompt engineering, agentic frameworks, and LLM orchestration (e.g., multi-step chains, tool use, structured output extraction).”
The ad says“Comfortable building and consuming MCP servers to integrate internal tools, data, and workflows into your dev loop.”
The ad says“Proficient in Python, FastAPI, PostgreSQL, containerization and scaling microservices.”
The ad says“Proficient in Python, FastAPI, PostgreSQL, containerization and scaling microservices.”
The ad says“Strong fundamentals in RAG architectures, prompt engineering, agentic frameworks, and LLM orchestration (e.g., multi-step chains, tool use, structured output extraction).”
The ad says“Proficient in Python, FastAPI, PostgreSQL, containerization and scaling microservices.”
The ad says“Strong fundamentals in RAG architectures, prompt engineering, agentic frameworks, and LLM orchestration (e.g., multi-step chains, tool use, structured output extraction).”
The ad says“Comfortable in (or excited to ramp on) TypeScript, React, and modern frontend tooling. You don't need to be a CSS wizard on day one, but you need to be the kind of engineer who'll learn React deeply rather than punt it to someone else.”
The ad says“Comfortable in (or excited to ramp on) TypeScript, React, and modern frontend tooling. You don't need to be a CSS wizard on day one, but you need to be the kind of engineer who'll learn React deeply rather than punt it to someone else.”

Values

The ad says“Deep understanding of transformer architectures, attention mechanisms, and GPU inference — you can reason about why a model behaves the way it does, not just prompt around it.”
The ad says“Experience with AWS (Bedrock, SageMaker, or similar managed AI infra), Supabase, LaunchDarkly & Betterstack.”
The ad says“Experience with AWS (Bedrock, SageMaker, or similar managed AI infra), Supabase, LaunchDarkly & Betterstack.”
The ad says“Experience with AWS (Bedrock, SageMaker, or similar managed AI infra), Supabase, LaunchDarkly & Betterstack.”
The ad says“Familiarity with AI observability and tracing tools (LangSmith, Braintrust, or similar).”
The ad says“Experience with fine-tuning, RLHF, or distillation workflows.”
The ad says“Experience with vector databases, embedding models, and retrieval optimization (hybrid search, reranking, chunking strategies).”
The ad says“Experience with fine-tuning, RLHF, or distillation workflows.”
The ad says“Deep understanding of transformer architectures, attention mechanisms, and GPU inference — you can reason about why a model behaves the way it does, not just prompt around it.”
The ad says“Familiarity with AI observability and tracing tools (LangSmith, Braintrust, or similar).”
The ad says“Experience with AWS (Bedrock, SageMaker, or similar managed AI infra), Supabase, LaunchDarkly & Betterstack.”
The ad says“Experience with vector databases, embedding models, and retrieval optimization (hybrid search, reranking, chunking strategies).”
The ad says“Experience with fine-tuning, RLHF, or distillation workflows.”
The ad says“Experience with AWS (Bedrock, SageMaker, or similar managed AI infra), Supabase, LaunchDarkly & Betterstack.”
The ad says“Experience with AWS (Bedrock, SageMaker, or similar managed AI infra), Supabase, LaunchDarkly & Betterstack.”
The ad says“Deep understanding of transformer architectures, attention mechanisms, and GPU inference — you can reason about why a model behaves the way it does, not just prompt around it.”
The ad says“Experience with vector databases, embedding models, and retrieval optimization (hybrid search, reranking, chunking strategies).”

Alma is an AI-native law firm. We combine AI with an exceptional legal team to deliver fast, transparent, and high-touch legal services at a fraction of the cost and speed of legacy firms. Our first market is immigration law. We're focused on simplifying complex immigration processes for companies and individuals, and disrupting the $15B market that has never seen an amazing product.

We are backed by leading VCs such as Bling Capital, Forerunner, Village Global, NFX, Conviction, MVP Ventures, NEA and Silkroad Innovation Hub.

🧠 Founding Team

The founding team has an extensive background in the legal industry, consulting, and building ML platforms.

Shuo - CTO & Cofounder: Shuo built the ML Platform (Michelangelo) at Uber for 5+ years after which he was the head of AI & ML team at a SupportLogic (a series B startup using predictive & generative AI for customer support).
Aizada - CEO & Cofounder: Aizada is a graduate of Harvard Law School, an attorney with 7+ years of experience including working at a top law firm such as Cooley and an ex-McKinsey consultant.

🔍 What we're looking for

We're hiring a Product Engineer (AI) — an engineer who lives at the intersection of AI systems and user-facing product. You'll own features end-to-end, from designing agentic pipelines and RAG architectures to building the React interface the user interacts with. Your north star is the product: you measure your work by whether it makes customers' lives better, not by model benchmarks in isolation. You'll work directly with the founding team and early customers to ship a world-class AI-powered immigration platform and help us scale by 100x.

You need to be deeply customer-obsessed. At Alma, "customer" means two groups. The first is our external customers — companies and individuals navigating immigration under high-stakes circumstances. The second is our internal legal team — attorneys and paralegals who use our platform every day. If an AI-generated document hallucinates a citation, that's not an ML problem — it's a customer problem. If an agent workflow saves a paralegal two hours a day, that's a feature worth shipping yesterday. The best engineers we'll hire will sit with the legal team, watch how they interact with our AI tools, and obsess over closing the gap between what the model produces and what the attorney actually needs.

At Alma, the engineer who designs the agent also builds the UI that exposes it. You should be the kind of engineer who can design a multi-step agentic pipeline in the morning, debug a prompt regression after lunch, and wire up the frontend that lets a user review the output before dinner. With modern AI tooling, the traditional excuse of "frontend isn't my strength" no longer holds.

You also need to build with AI, not just build AI. You should already be living inside modern coding harnesses (Claude Code, Codex, Cursor, etc.), wiring up MCP servers to extend your own workflow, and treating agents as a core part of how you ship. If your instinct is "let me spin up an agent to scaffold this, then review and harden it," keep reading.

You see problems as work to do, not things to complain about. We move fast — our codebase has rough edges, our processes are still being invented, and not every corner is beautiful yet. We want engineers whose instinct, when they hit a messy module or a flaky eval, is "I'll fix it" — not "who wrote this?" If you're the kind of engineer who quietly improves the codebase around every feature you ship, you'll thrive here.

✅ Responsibilities

Work closely with the founding team and early customers to build a world-class AI-powered product.
Own AI-powered features end-to-end — design the agentic pipelines and LLM integrations, build the backend services that orchestrate them, create the frontend that exposes them, and ship to production.
Design, build, and iterate on RAG pipelines, agentic workflows, prompt chains, and evaluation frameworks that power our core product.
Develop and maintain robust evaluation and observability systems — you'll define how we measure whether our AI outputs are actually good, and build the tooling to catch regressions before customers do.
Partner with our designer and product team to translate user needs into shipped AI-powered experiences — not just APIs or model outputs that "someone else will integrate later."
Operate as a force multiplier with AI tooling — use coding agents to move 5–10x faster than traditional engineers without sacrificing quality, and continuously evolve how our team builds, reviews, and tests software.
Engage in the entire application lifecycle with a high bar for coding, debugging, testing, and deploying.

🛠 Role Requirements

2+ years of experience building production AI/LLM-powered features — not research prototypes, but systems that real users depend on. You've dealt with hallucination mitigation, latency constraints, cost optimization, and the messy reality of shipping AI to production.
Strong fundamentals in RAG architectures, prompt engineering, agentic frameworks, and LLM orchestration (e.g., multi-step chains, tool use, structured output extraction).
Experience designing evaluation and quality systems for AI outputs — you've built evals, human-in-the-loop review flows, or regression detection pipelines, and you have opinions about what "good enough" means for production AI.
Proficient in Python, FastAPI, PostgreSQL, containerization and scaling microservices.
Comfortable in (or excited to ramp on) TypeScript, React, and modern frontend tooling. You don't need to be a CSS wizard on day one, but you need to be the kind of engineer who'll learn React deeply rather than punt it to someone else.
Daily, hands-on use of modern coding harnesses (Claude Code, Codex, Cursor, or equivalent) — not as autocomplete, but as a primary way of shipping production code.
Comfortable building and consuming MCP servers to integrate internal tools, data, and workflows into your dev loop.
Strong point of view on AI-assisted code review, automated testing, and how to keep quality high when shipping at agent speed.
Enjoy working in a fast-paced environment & wear multiple hats — including ones outside your historical comfort zone.
Ability to seek & find the right resources for solving open-ended problems.
Located in the San Francisco Bay Area or willing to relocate.
BS/MS in Computer Science, Engineering, or a related technical field.

⭐️ Nice-to-haves

Experience working in a small startup environment (Seed or Series A).
Deep understanding of transformer architectures, attention mechanisms, and GPU inference — you can reason about why a model behaves the way it does, not just prompt around it.
Experience with fine-tuning, RLHF, or distillation workflows.
Experience with vector databases, embedding models, and retrieval optimization (hybrid search, reranking, chunking strategies).
Familiarity with AI observability and tracing tools (LangSmith, Braintrust, or similar).
Experience with AWS (Bedrock, SageMaker, or similar managed AI infra), Supabase, LaunchDarkly & Betterstack.
Built your own MCP servers, custom agents, or internal AI dev tooling.
Published, blogged, or shared opinions about modern AI-assisted development workflows.
Genuine full-stack experience — comfortable across backend, frontend, and infra.

Benefits & Perks

Hyper-competitive base salary range based on experience
100% Health coverage + Vision + Dental
20 days PTO + 10 Federal Holidays
$250 per month in wellness allowance