Job Title / Role: Senior AI Engineer

Type of Employment: Full Time & C2C

Years of Experience Required: 3–9 yrs

Number of Positions: 6

CTC in INR: 30–60 LPA

Work Location: Bangalore (Preference given to Bangalore) Kothanur

Mode of Work: Remote

Notice Period / Start Date: 0–30 days (Serving Only)

In office feasibility (work abroad): Yes

Interview Rounds: L1 - Assessment Q&A;, L2- Tech, HR

Office Time: 9:30 AM to 6:30 PM

The Project

We're building a multi-agent orchestration platform for enterprise document and transaction processing. The technical core is durable orchestration — long-running workflows that pause for human approval for hours or days, recover from tool failures, and replan without losing state. Agents read unstructured documents, classify against complex taxonomies, and write into legacy systems through typed connectors spanning mainframe terminal emulation, EDI, and JDBC.

Everything is versioned and replayable: agent manifests, prompts, eval scores, and an immutable decision ledger that lets you reconstruct why any action was taken.

Identity is first-class. Every agent runs as its own authenticated service principal, not a shared credential, with scoped entitlements per connector method and per client. Access is brokered just-in-time on short-lived leases — agents never hold long-lived keys.

Nine business domains, roughly forty agents, phased from shadow mode to supervised autonomy.

Roles & Responsibilities

- Build and own sub-agents: typed input/output schemas, versioned prompts,



a constrained tool set, and a published success metric for each.

- Design the orchestration layer — planning, replanning, tool-failure recovery, and handoff between supervisor and sub-agents.

- Build the eval harness that decides when an agent is safe to promote from shadow to assisted to autonomous.

- Work the two-tier model setup: a heavy model for judgment and classification, a lighter self-hosted tier for high-volume extraction and formatting.

- Make agent behaviour observable and debuggable — traces, cost per agent, failure attribution.

Must-Have Details

- You've shipped LLM systems into production, not just demos. You can talk about what broke and what you changed.

- Comfortable with agent orchestration in some form — LangGraph, custom runtimes, workflow engines, or your own. We care more about the reasoning than the library.

- Strong Python. Go or TypeScript useful.

- You've built evals, or you've felt the pain of not having them.

- Solid on structured output, schema validation, and getting reliable behaviour out of unreliable models.

Good-to-Have Details

- Durable execution experience — Temporal, Cadence, Step Functions, or similar.

- Self-hosted inference: vLLM, TGI, quantisation, GPU scheduling.

- Retrieval systems at scale, and knowing when not to use them.

- Document understanding: OCR, layout parsing, extraction from messy PDFs.

- Anything regulated — audit trails, approval workflows, compliance reporting.

📌 Senior AI Engineer — Multi-Agent Platform (Bengaluru)
🏢 HyrEzy Talent Solutions
📍 Bengaluru