Lead AI Engineer | Enterprise Agentic Systems

Overview

About Atherna
Atherna is a Legal AI company redefining how lawyers work. We build intelligent systems for leading law firms, in-house legal teams, financial institutions, and banks — spanning document analysis, drafting, research, and matter management. We're moving fast and building something that genuinely changes how an entire profession operates, and we're already trusted by leading enterprises across India.
This is a full-time engineering role on the Atherna product team. You'll work directly with the founders, shipping production systems that real enterprise clients depend on — from day one.

Our Benchmark: this role is about agentic systems — autonomous, multi-step software that does real legal work, not chat demos. The problems below are the actual job, not résumé keywords. AI has made writing code faster than ever and dropped the floor for the whole industry; producing something that runs is nearly free now. When the easy 80% is cheap, the value is in the part AI can't hand you — architectural judgment, the right abstraction, the reusable interface, the decision not to build something at all. We hire for that judgment, not typing speed. If you've wired up a handful of API calls but haven't built and operated agents in production for real users, this will be a stretch for you. If you have, you'll recognize every line below.

This is real ownership at a fast-growing startup — high stakes, quick pace, and meaningful upside. It also means there are stretches, around launches and enterprise deadlines, where the work runs past standard hours. We want people energized by that, not worn down by it.

What You'll Do

• Own full-stack features end to end. Next.js frontend, Node.js/Python backend. You own the feature from database to UI, not just your slice of it.
• Build agentic systems. Write the orchestration that lets multiple agents collaborate on multi-step legal work — planning, drafting, analysis, research — coherently and without human babysitting. You'll design planner/sub-agent patterns, tool use, and control flow, and fight the real enemies: incoherence, repetition, drift, and silent failure.
• Design and ship custom AI SDKs. Own our internal developer primitives — robust, type-safe Python and TypeScript SDKs with clean interfaces that let agents, internal teams, and enterprise clients work with our core orchestration and tool protocols.
• Build for reuse, not for the demo. We ship customized software to every client on a shared modular architecture, not a forked codebase per customer. Your job includes growing that architecture so a new client's requirements drop in through configuration and composition, not a from-scratch rebuild — clean seams between modules, and the judgment to know where an abstraction earns its place and where forcing one just adds coupling.
• Implement secure sandbox code execution. Legal agents parse massive data structures, run complex financial calculations, and execute dynamic Python on the fly. You'll design and maintain isolated, secure sandboxes (microVMs, WASM, or containerized runners) so agents run arbitrary code safely without risking host infrastructure.
• Make automation reliable end to end. Real clients trust these systems with real work, so the whole pipeline — input to verified output — has to be dependable: retries, guardrails, output verification, and safe failure when a step goes wrong.
• Engineer RAG that holds up under scrutiny. Retrieval quality, citation integrity, hallucination mitigation, and calibrated "I don't know" behavior. When the system declines too often — or confidently invents things — you know what to instrument and what to change.
• Make quality measurable. Build eval harnesses for the systems you ship. A/B test prompt changes without breaking users. Treat "5% bad answers" as a number you can move, not a vibe.
• Operate agents in production. Structured logging and tracing across every step and tool call, observability into multi-agent runs, latency and cost control, and calm incident response when something regresses at 2am.
• Secure systems for enterprise. Defend against prompt injection — a pasted document should never be able to exfiltrate a conversation. Manage secrets and provider API keys properly. Reason about network isolation (VNet/VPC), database hardening, and locking down internet-facing storage.
• Design data for scale. Model multi-tenant data in PostgreSQL — workspaces, users, projects, chats — and build pgvector search that stays fast and correct as the corpus grows.

Who We're Looking For
We care far more about what you've shipped than where you've worked or for how long. Proof of work does the talking: shipped products, live systems, automation people actually rely on. Strong data-structures-and-algorithms fundamentals matter, but they're the floor, not the differentiator; clean LeetCode on its own won't land this role. Competitive hackathon wins count, weighted below production work real people depend on. If you're early in your career but have genuinely built and operated agentic systems in production, we want to talk. If you have 2–3 years and a track record of owning things end to end, even better.

What you need:

• Engineering judgment, not just output. You make good structural decisions under real constraints — when to abstract and when to leave it inline, how to shape a function or class so the next engineer (and the next client) reuses it instead of rewriting it, how to grow a modular codebase without it rotting into spaghetti. AI can generate code; it can't make these calls for you.
• Backend depth & SDK craftsmanship. Strong Node.js or Python. You don't just consume API wrappers; you design, package, and distribute ergonomic, production-grade SDKs in both Python and TypeScript, with strict type-safety, clean error-handling, and intuitive interfaces.
• Frontend craft. React/Next.js. You structure components in a large app without leaving a mess for the next person.
• Production agentic experience. You've built agents that do real work — tool use, multi-step orchestration, sub-agents — on frontier model APIs. You understand prompt chaining, context management, RAG, evals, and where agents break under real load.
• Sandbox & isolation experience. You know that letting an LLM generate and run its own code is powerful but volatile. You have a sharp instinct for runtime boundaries, strict resource limits (CPU/memory), input/output serialization, and keeping execution strictly non-persistent.
• Agent infrastructure literacy. You know what an agent harness is, the tradeoffs between frameworks, and where MCP and agent "skills" fit. You've thought hard about memory across long conversations and what to do when an agent has more tools than fit in context.
• Security instinct. You think about how something gets abused before you ship it. Secrets management, prompt-injection controls, and the fundamentals of network, database, and storage hardening are reflexes, not afterthoughts.
• Cloud comfort. You can touch AWS, Azure, or GCP without flinching — we're not religious about which. You don't need to be a DevOps engineer; you just can't be afraid of one's job.
• Ownership. You push things to done — and "done" means reliable in front of real users, not a green checkmark on a ticket. You raise blockers early, iterate fast on feedback, and care about the outcome.
• An honest AI-assisted workflow. You use AI coding assistants as part of how you build, and you can explain exactly how — what you delegate, what you don't, and why.

How We Interview
No pure LeetCode gauntlet — solid fundamentals are assumed, but they aren't what we're testing for. Expect us to take something you've actually built and keep pulling the thread: why this architecture and not another, what broke in production, what you measured, what you'd do differently now. Come ready to think out loud and defend real decisions, not to recite.

Tech Stack
Frontend: Next.js, React, TypeScript
Backend & Tooling: Node.js, Python, custom TS/Python SDKs
Data: PostgreSQL + pgvector; SQL/NoSQL as appropriate
AI / Agents: Agentic orchestration, MCP, RAG pipelines, frontier model APIs
Execution & Security: Isolated code sandboxes (Docker / WASM / microVMs)
Cloud: AWS, Azure, or GCP (cloud-agnostic)

What You'll Get

• Real ownership. Features, not tasks. Production systems, real enterprise users, actual stakes.
• Direct founder access. Work closely with the CEO and CTO on architecture, product, and technical strategy.
• Accelerated growth. Hard problems and a tight feedback loop — you'll grow more in six months here than in two years most places.
• Founding engineer path. Strong performers move toward a founding-engineer designation with meaningful equity and leadership scope.
• Long-term upside. Equity, ownership, and a seat at the table as we scale.

Details
Type: Full-time
Location: On-site (Bengaluru)
Compensation: Competitive salary + equity (commensurate with experience)
Start date: Immediate

How to Apply
Email careers@atherna.ai with subject "Lead AI Engineer Application - [Your Name]" and include:

• Your resume. If you've been one of the early people a startup's progress genuinely depended on — not just an employee — call that out explicitly.
• A link to your GitHub (or other public code).
• A short note on your work — real things you've shipped to users (live links/demos strongly preferred), open-source projects with traction (downloads, stars, adoption), enterprise customers you've built for and roughly how many, the genuinely geeky hard things you've built for the love of it, and any competitive hackathon wins or side projects that show how you think and build. We care deeply about proof of work. Show us what you've built — and how you think.