AI Engineer, Agentic Systems, Lawyered Location: Gurgaon (Hybrid) | Experience: 3–6 years | Type: Full time Company: Lawyered, part of the India Accelerator ecosystem About Lawyered Lawyered is building AI-native legal help for India. We're moving beyond chatbots that answer questions to AI agents that act: they take a user's legal problem from intake to resolution by researching the law, drafting documents, filing on government portals, tracking cases, and escalating to human lawyers when judgment is needed. The Role You'll architect and ship autonomous and semi-autonomous agents that do real legal work end-to-end. That means designing how agents plan, use tools, remember context, recover from failure, and know when to hand off to a human. In a regulated domain, reliability, auditability, and safety are core to the job, not afterthoughts. What You'll Build

- Multi-agent legal workflows. You'll orchestrate specialized agents (intake, research, drafting, filing, follow-up) that collaborate on a single matter, such as a consumer complaint moving from facts to legal notice to e-Daakhil filing to hearing reminders.
- Tool-using agents. You'll give agents secure access to tools and APIs via function calling and MCP, including legal databases, eCourts, consumer forums, DigiLocker, payments, e-sign, WhatsApp, and email.
- Browser and computer-use agents. You'll automate interactions with government and court portals that lack APIs, with robust handling of CAPTCHAs (via human handoff), timeouts, and UI changes.
- Long-running, stateful agents. You'll use durable execution (Temporal, LangGraph, or similar) for matters that span weeks,



with checkpointing, retries, and resumability.
- Agent memory. You'll design short-term, episodic, and case-level memory so agents retain facts, documents, and history across sessions without leaking context between users.
- Planning and self-correction. You'll implement task decomposition, reflection loops, and verifier agents that check citations, facts, and document validity before anything reaches a user.
- Human-in-the-loop controls. You'll build approval gates for high-stakes actions (filing, sending notices, payments), plus confidence-based escalation to in-house lawyers and seamless lawyer takeover.
- Grounded legal reasoning. You'll build agentic RAG over statutes, case law, and regulations, where agents decide what to retrieve, iterate on queries, and cite every claim.

What You'll Own
- Agent evals. This covers trajectory-level evaluation (did the agent take the right steps, not just produce the right answer), task success rates, tool-call accuracy, and regression suites built with lawyers.
- Observability. You'll trace every agent decision, tool call, and handoff, keeping full audit trails suitable for legal accountability.
- Safety and permissions. You'll scope tool access tightly, sandbox execution, defend against prompt injection from user uploads and web content, and enforce action-level authorization.
- Cost and latency. You'll handle model routing across frontier and open-source models, caching, and token budgets per matter.
- Compliance. You'll ensure DPDP Act–compliant data handling, including consent, PII redaction, and retention policies across agent memory.

📌 AI Engineer (Gurugram)
🏢 Sproutech Solutions Private
📍 Gurugram