About This Role

- Ship end-to-end AI-powered workflows frontend, backend, agent logic, evals as one person, not as a relay race across three teams.

- Build multi-step agentic workflows that retrieve documents, use tools, and complete drafting, review, and research tasks a paralegal or associate would otherwise do by hand.

- Wire up LLMs, RAG pipelines, and document-processing chains against real legal source material not toy datasets and adapt them to each firm's practice area and working style.

- Put guardrails around everything: evals, monitoring, fallback paths, and a human-review step for anything that touches legal risk. No silent failures in front of a client's attorney.

- Own token economics like a founder owns burn routing, caching, batching, and picking the right model for the right step, because solo and small-firm pricing only works if delivery cost stays low.

- Use coding agents and AI-assisted dev tools as your default way of writing software we expect you to move faster because of AI, not despite the extra tooling.

- Sit close to attorneys and legal engineers so the workflows you build reflect how legal work actually happens at a small firm, not how engineers assume it happens.

What gets you shortlisted

- 57 years shipping production software full-stack you've owned something end-to-end that real users depended on, not just tickets in a sprint.

- Comfortable in a modern frontend stack (React / Next.js / Vue) and a backend stack (Python / Node.js / Java / .NET) pick your poison, but be genuinely robust in at least one of each.

- You've actually built something with LLMs in production an agent, a RAG pipeline,



a document intelligence system and can talk about what broke and how you fixed it, not just what the architecture diagram looked like.

- Solid instincts on system design, security, and performance — you know when a deterministic script beats an agent, and you're not afraid to say so.

- You already use AI coding assistants / agents as part of how you write code day to day.

- APIs, databases, cloud (AWS/GCP/Azure), containers, CI/CD, production monitoring — the boring fundamentals that keep 2am pages from happening.

Extra credit

- You've optimized inference cost for real — model routing, caching, batching, or built the eval harness that told you it was safe to switch models.

- Fine-tuning, post-training, or continual learning experience.

- Hands-on with LangGraph, LlamaIndex, Semantic Kernel, AutoGen, CrewAI, or you've rolled your own agent framework because the existing ones didn't fit.

- Vector DBs, open-source model deployment, or running models locally/on-prem.

- Legal tech, practice management software, contract review, document intelligence, or small-firm/solo practitioner workflow automation — you already speak some of the domain language.

How we think about this role

We're not looking for someone who waits for a spec. We're looking for someone who can sit with an attorney for twenty minutes, understand the actual pain, and come back with working software — not a Jira epic. You should have a strong, opinionated view on where AI genuinely helps versus where a plain rules engine is just better, cheaper, and more auditable. Say that out loud in the interview; we'll take it as a good sign, not a red flag. Proces

📌 AI-Native Full Stack Engineer (Bengaluru)
🏢 LegalEase solutions
📍 Bengaluru