WaterDoctor — the AI agent stack behind biological water treatment.
WaterDoctor is a Singapore deep-tech company founded by NUS PhD researchers in environmental biotechnology, microbial ecology, and AI. It builds AI-integrated biofilm reactors for water treatment and aquaculture. wGrow is a strategic partner and co-investor — we designed, built, and operate the agent crew that runs WaterDoctor's adaptive treatment loops in production.
Across pilot tanks vs prior baseline.
Far less make-up water needed per cycle.
Adaptive aeration over fixed schedules.
Outcome figures reported by WaterDoctor; methodology and pilot conditions referenced in their public communications at waterdoctor.com.sg.
The relationship
We disclose this up front because it shapes everything else on this page. wGrow is not a vendor that delivered a project to WaterDoctor. wGrow is a strategic partner with capital in the company, an ongoing technology partnership, and a shared roadmap. Our embedded delivery crew works inside the WaterDoctor team — same standups, same eval harness, same production rota. When we say we ship our agent crews into real systems, this is the engagement we mean.
The problem
Treating water with microbial biofilms is a continuous, noisy process. Sensor readings drift, biology adapts, weather shifts loadings. Static control schedules over-aerate when they should rest, recirculate when they should hold, and chase symptoms while the underlying community shifts underneath. WaterDoctor wanted a controller that would learn the local process — and a way to prove to a regulator that the controller was behaving.
The crew shape
Four narrow agents. Two senior wGrow engineers. The agents are ingestor
(telemetry + lab assays), controller (treatment decisions
inside engineer-defined safety envelopes), reporter
(regulator-readable run summaries grounded in raw data), and eval
(continuous deviation monitoring). The engineers own the safety envelopes, the schema, and the
sign-off on every regulator-bound report. WaterDoctor's process scientists own the biology;
we own the agent stack and the audit trail.
What we got right
- · Narrow agents. No agent had a job description that read like a management role.
- · Hard safety envelopes. The agent picks moves; the engineers wrote the floor and ceiling.
- · Reports that link back to raw data. No paraphrased numbers ever leave the system.
- · An eval harness from week one. Drift is a first-class signal, not a postmortem.
- · Co-located decision-making. WaterDoctor's process scientists sit a corridor away from our crew.
What we got wrong (and changed)
- · Tried a single "ops agent" first. Too broad. Split into ingestor / controller / reporter / eval after week three.
- · Underweighted lab-assay latency. Lab values arrive days after a sensor reading. Controller assumed alignment. Fixed with a delayed-supervision channel.
- · Auto-tuned aeration too aggressively in week six. Engineer-written guardrails caught it; we tightened the envelope and added a human-approval gate for any change >15% from rolling baseline.
Why this is the engagement we lead with
Most public agentic work is chatbot demos. This one moves real molecules in real tanks under real regulatory eyes, and the outcomes are measured by an environmental engineer, not a marketer. That's the kind of work the rebrand is for. And because we have skin in WaterDoctor — capital, shared roadmap, joint reputation — we can stay honest about what works and what doesn't.