AI Solutions Architect / Applied AI Engineer

About HySights

HySights maps, tracks, and connects pre-production clean fuel projects globally, covering bio-methane, SAF, bio-LNG, ammonia, e-methanol, and green hydrogen. Our proprietary project and deal data is accessed by leading global institutional investors, offtakers, and advisors through a subscription WebApp. We are live on London Stock Exchange Group's (LSEG) Workspace and backed by EDB Singapore..

We are building a conversational AI feature that allows subscribers to query our proprietary data in natural language, the way they would brief and interact with an energy analyst. 

The Role

You will be responsible for building the HySights conversational AI feature, Ask HySights AI, end to end: the architecture (an orchestration microservice, the MCP tool layer that maps subscriber questions to structured queries against our Directus API, and Bedrock guardrails), the quality of the answers, and the decisions that shape it and the ongoing evolution of its output, i.e. from discover capability, to analytic capability, to narrative-based output. We currently work with an outsourcer whom we envision will continue to play a role in helping with parts of the build, for example the chat UI, connector work, or testing.

You will work directly with the senior data operations leader, a domain expert who maps our data, builds templates, encodes analytic frameworks into the underlying knowledge base your evaluation benchmarks draw on. 

Two further areas are also open for discussion as this role takes shape. WebApp work, including loading time and general improvements, is currently outsourced. For IT hygiene, we are open to exploring an outsourcer as well. We'll work through the right split with you directly (see below).

What You Own

•     Architecture and build: the agent, the MCP tool layer, and how a subscriber question maps to a structured Directus query (text-to-query). We are open to discussion on the scope of role an outsourcer plays in helping with parts of this.

•     Evaluation criteria and quality benchmarks: build and maintain a set of real questions with known-correct answers, run every skill against it before launch and after every Bedrock or Claude model update.

•     Hallucination rate tracking and reduction, and the call on what is good enough to ship.

•     Security and data handling: server-side paywall enforcement, read-only least-privilege access to Directus, prompt-injection defence, audit logging, and GDPR readiness for European subscribers.

•     Specs precise enough for any additional development support to build against without ambiguity, when you bring some in, plus review of their output against those specs.

•     Reviewing handover material (architecture diagram, maintenance guide, plain-language brief) at the end of each build phase, and validating it is sufficient to operate and extend the feature independently.

•     Escalating to the senior leadership with clear options, not just problems.

•     Designing the system so later additions (the digest feature, the remaining tiles, the Analysis and Narrative layers, new personas) sit on the same foundation without a rebuild.

•     IT hygiene: device management, IT security, and how the app runs and operates day to day. We are open to exploring an outsourcer for this; in discussion as this role takes shape.

•     WebApp: improvements to the WebApp, including loading time. Currently outsourced; open to discussion on how this is resourced going forward.

What We Are Looking ForMust Have

•     Production experience building or maintaining at least one LLM application.

•     AWS Certified Solutions Architect – Professional, AWS Certified DevOps Engineer – Professional, or an AWS Specialty certification (Security, Advanced Networking, or Machine Learning).

•     Ability to explain RAG in plain English, articulate how it differs from fine-tuning, and judge when a direct text-to-query approach over structured data is the better fit, as it is for our roughly 1,000-row dataset. If you cannot explain it to a non-technical CEO, this role is not for you.

•     Hands-on AWS experience, Bedrock preferred. Python or JavaScript/TypeScript.

•     Strong technical writing, for the times you bring in extra development support and need to spec precisely what you want built.

•     Judgment and the confidence to challenge flawed requirements, including proposals that add more infrastructure than a small structured dataset needs (a vector database, a data warehouse, fine-tuning). This is a small team with a non-technical CEO. We need someone who pushes back when something is wrong.

•     Comfort building hands-on and, when needed, managing additional contracted development support: reviewing pull requests and giving specific, testable feedback against a spec.

Nice to Have

•     Directus or headless CMS experience.

•     Prior chat UI build experience.

•     Experience managing external or contracted development teams, including vendors responsible for IT security, device management, or general web application performance.

•     Energy, commodities, or financial data domain background.

•     Familiarity with agentic workflows, tool calling, and MCP implementations (e.g. Claude Code, Hermes, Codex).

What We Offer

•     Base salary: S$10,000 to S$12,000 per month, commensurate with experience. We are hiring at mid level, not senior.

•     A small, high-quality team with significant institutional backing (EDB Singapore, IMI plc, Rainmaking APAC).

•     Live distribution on LSEG Workspace. Your work has immediate reach into institutional markets from day one.

•     Direct access to the CEO and co-founders. No layers.

•     The opportunity to build something that, if successful, becomes infrastructure for how clean fuels capital is deployed globally.

Experience Level

This field is new enough that most people do not have years of it. We are not looking for a senior hire. We are looking for someone who has built things with these tools, understands the AWS environment, and has the judgment to know which architectural decisions matter. A few years of relevant experience is the right bar. Older with more experience is equally welcome.

How to Apply

Send your CV with a cover note to [Confidential Information] with a short description of something you have built using LLM infrastructure: what you used, what decisions you made, and what you learned.