AI Architech

Meeru AI Inc; United States; À distance; USD 180 000 - 240 000; Plein temps

Résumé du poste

Meeru AI Inc in the United States is seeking an AI Architect to design and build the intelligence engine at the core of our platform, and the conversational agents that sit atop it to deliver deterministic, auditable answers. The role emphasizes grounding, evaluation rigor, and a seamless integration with customer data through trusted AI workflows.

This is a deeply hands-on architecture role, requiring leadership of engine and agent layers, collaboration with US leadership, and a strong focus on

Qualifications

• Shipped production AI systems with deterministic outputs.
• Experience leading technical direction on at least one production AI system.

Connaissances

Shipped production AI systems

LangGraph or similar framework

Retrieval-augmented generation (RAG)

Tool-using agents

Structured output (JSON schemas, func.

Agent orchestration

Python production software

SQL access to warehouses (Snowflake,PG

Containerized deployment

Knowledge graphs and reasoning engines

Grounding and eval in CI

Determinism with probabilistic models

Outils

LangGraph

pgvector

Pinecone

Weaviate

Qdrant

Snowflake

Description du poste

About Meeru AI

Meeru AI is building an AI-native platform that transforms how finance and accounting teams operate. We connect to enterprise financial systems — ERPs, CRMs, billing platforms, HRIS — and apply machine learning to turn fragmented operational data into grounded, auditable intelligence for CFOs, controllers, and FP&A leaders.

We deploy on customer terms — SaaS multi-tenant, SaaS single-tenant, and on-premises — across AWS, Azure, and GCP. Our customers are Fortune 500 finance teams who require data isolation, auditability, and compliance.

The Role

We are looking for an AI Architect to help design and build the intelligence engine at the core of our platform, and the conversational AI agents that sit on top of it and serve answers deterministically.

Our output sits adjacent to externally reported financials. “Sounds plausible” is never good enough. The engine you help architect turns connected enterprise data into answers that are computed, reproducible, and traceable to verified source data — not improvised by a model. The agents that sit on top interpret what a user is asking, route it to the engine, and present the result in natural language without ever inventing the substance of the answer.

This is a deeply hands-on architecture role. You set technical direction for the engine and agent layer, partner closely with US-based AI leadership and the Staff AI Engineer on grounding and evaluation rigor, and raise the bar for how deterministic, auditable AI gets built across the team.

What makes this role different

Deterministic by design — the engine computes answers from verified data. The same question against the same data returns the same grounded, explainable answer.

A clear separation of concerns — the engine reasons and decides; the conversational agents orchestrate and narrate, but never fabricate the substance.

Auditable end to end — every answer traces back to its source data and the logic that produced it, because our customers’ auditors will ask.

Deploys in customer clouds — including managed LLMs in-VPC, so customer data never leaves their environment.

• Senior or staff-level applied AI/ML experience. You have shipped production AI systems, not just prototypes, and have led the technical direction on at least one of them.
• Deep fluency in modern LLM patterns. Hands-on experience with retrieval-augmented generation (RAG), tool-using agents, and structured output (e.g., JSON schemas, function calling).
• Agent orchestration. You have built multi-step agent workflows with LangGraph or a similar framework, including verification steps, fallbacks, and points where a human confirms before the system proceeds.
• Systems architecture. You can design an engine, its data interfaces, and the agent layer on top as one coherent, maintainable system, and document decisions (e.g., architecture decision records).
• Grounding, hallucination control, and evaluation. You have made “no unsupported answers” a hard rule in a real system, and measured faithfulness and hallucination with automated evaluations, ideally running in CI.
• Determinism on top of probabilistic models. You know how to make the same question against the same data return the same answer, and can explain the techniques you used.
• Clear judgment about what a model can legitimately claim. You understand why computed, traceable answers matter more than plausible-sounding ones in a high-stakes domain.
• Production software engineering in Python. You write robust, tested services with SQL access to warehouses or databases (e.g., Snowflake, PostgreSQL), and you are comfortable with containerized deployment.
• Structured data and retrieval tooling. Working knowledge of vector stores (pgvector, Pinecone, Weaviate, or Qdrant) and of computing reliable answers over structured data.
• Technical leadership. You mentor engineers, run design reviews, and turn strategy set by US leadership into well‑built execution.
• Strong communication. You can explain architecture trade‑offs to engineers, product managers, and non‑technical stakeholders in clear English.
• Location and working hours. Based in Hyderabad (preferred) or remote elsewhere, and able to keep a few hours of daily overlap with US time zones.
• Flexible engagement. Open to full‑time, contract, or contract‑to‑hire, whichever suits you best.
  Nice‑to‑Haves
• Finance, accounting, or other audited domains. Experience where outputs face auditors or regulators (finance, healthcare, insurance, legal).
• Knowledge graphs and reasoning engines. Experience with knowledge graphs, semantic layers, or rule/reasoning engines that model facts and relationships.
• Human‑in‑the‑loop systems. You have designed review, confirmation, or feedback loops that improve a system over time.
• Managed LLMs inside customer clouds. Hands‑on with Amazon Bedrock, Google Vertex AI, or Azure OpenAI deployed in a customer’s private network (VPC), so data never leaves their environment.
• Multi‑cloud and on‑prem deployment. Experience shipping AI to AWS, Azure, and GCP, including single‑tenant and on‑premises installs, and working with platform teams on model serving and inference cost.
• AI observability. Experience with tracing and evaluation tooling for LLM applications.