Two boxes sit on a whiteboard. The left box says AWS Bedrock, the right box says Azure OpenAI, and there is no arrow yet. Most remote GenAI solutions architect jobs are really about what happens before that arrow gets drawn. This opening, full-time and remote with no country restriction, pays USD 125,000 per year to a senior architect who designs enterprise retrieval-augmented generation (RAG) platforms and can defend the choice of cloud underneath them.
Before choosing, the architect asks a few plain questions and writes the answers beside each box.
Data handling is part of this list too, but the architect checks the provider's current documentation and the client's own contract terms rather than repeating marketing lines from memory.
The platform in this role is a single enterprise RAG service that several business units will share. Functional requirements come first: answer questions over policy documents, cite the source passage, refuse when the answer is not in the corpus. Then the list that usually decides the design: availability targets, how fresh the index must be after a document changes, audit logging, data residency, accessibility of the chat interface, and what happens when the model provider has an outage. A sound GenAI solutions architect job description lists them beside features, since they drive more of the cost.
Described in words, the diagram has five lanes. On the left, source systems: file shares, a wiki, a ticketing tool. Next, an ingestion lane with connectors, a parser, a chunker and an embedding job that writes into a vector store. In the middle, a retrieval and orchestration service written in Python with LangChain. To its right, the model gateway that calls Bedrock or Azure OpenAI, wrapped in guardrails. On the far right, the client apps and an evaluation store that records every question, retrieved chunk and answer for review.
Chunking decisions are made per document type, not once for the whole estate. Policy PDFs with numbered clauses might split on headings with a small overlap; spreadsheets exported to text need their header row repeated in each chunk or the numbers lose meaning. Metadata such as owner, department, effective date and access group is attached at ingestion so retrieval can filter later.
The options fall into three broad groups: a vector index inside a search service the client already pays for, a vector extension on a relational database they already run, or a dedicated vector database. The architect weighs filtering support, hybrid keyword plus vector search, operational burden and how tenancy will be enforced, and writes the reasoning down so the next team understands why.
LangChain chains handle query rewriting, retrieval, reranking and prompt assembly. The architect sets conventions for prompt templates, retries, timeouts and tracing so that several engineering teams can add features without each one inventing its own pattern.
Guardrails sit on both sides of the model call: input checks for prompt injection and personal data, output checks for groundedness, topic limits and refusal wording. Evaluation is treated as a release gate. A labelled question set built with subject experts runs on every change to the chunker, the prompt or the model version, scoring retrieval hit rate, faithfulness to the retrieved text and answer quality. When a model upgrade improves fluency but starts citing the wrong clause, that set is what catches it before users do.
Several business units share one platform, so a finance analyst must never get an answer built from an HR investigation file. The design passes the user's identity and group claims from the client app through to retrieval, filters at query time on access metadata stored with each chunk, and keeps per-tenant encryption keys and logs where the client requires it. Secrets live in a managed vault, and every answer can be traced back to the documents a user was entitled to see.
Every request spends two things: tokens and time. The architect breaks each budget down by stage, so the team knows how much of the wait belongs to retrieval, reranking, the model call and post-checks. Trade-offs are explicit: a smaller model for query rewriting, caching for repeated questions, shorter context windows when the eval set shows no loss in quality, streaming responses so users see progress. Changes that push past a budget go back to design review.
You will lead proof-of-concept builds on AWS Bedrock or Azure OpenAI, then turn what was learned into decision records, a reference architecture document and a threat model. Solution designs are presented to client stakeholders, from security leads to business sponsors, in language each group can act on. Day to day, you review pull requests on the orchestration layer, pair with engineers on a tricky retrieval bug, and run a weekly design review where teams bring proposals and leave with a decision.
A bachelor's or master's in computer science or engineering is expected, along with 7+ years in software or cloud architecture, including recent hands-on LLM work. You should have deep skill in AWS Bedrock, Azure OpenAI, LangChain, vector databases and Python. An AWS or Azure professional-level certification is preferred, not mandatory. Strong GenAI solutions architect skills show up as shipped systems: a RAG service in production, an eval suite you designed, a migration you led.
Searches for entry level remote GenAI solutions architect jobs will not match this opening. It is a senior role, and the experience bar is part of the brief.
For this listing, the GenAI solutions architect salary per year is USD 125,000. The role is full-time and fully remote, one of the GenAI solutions architect jobs remote worldwide, so it also fits anyone looking for GenAI solutions architect work from home jobs. No other compensation details appear, so anything beyond the base figure is a question for the employer.
Expect GenAI solutions architect interview questions built around a whiteboard prompt: design a RAG assistant for a regulated client, then defend the cloud choice, the chunking plan and the tenancy model. Interviewers often follow with LLM evaluation: how you would build a test set, which metrics you trust, and how you would detect a regression after a model version change.
Coming from cloud architecture, the route into this kind of role is practical: build one end-to-end RAG project on each cloud, write a short decision record comparing them, and add an eval suite. Vendor documentation, free provider training and a GenAI solutions architect course from a reputable source can fill gaps. A GenAI solutions architect certification helps signal cloud depth, but working designs carry more weight.
Your GenAI solutions architect resume should lead with two or three systems you designed, the constraint that shaped each one, and what you would change now. Apply through the listing page on Naukri Mitra. Submitting an application costs nothing, and an employer hiring at this level has no reason to ask you for money at any point.