How to Train an AI — A Hands-On Workshop · Application: Wastewater Settling Level & Microbial Quality Labeling
You’ll be training an AI that monitors real wastewater plants. The images you label become production training data for models that help operators serving hundreds of thousands of people. Not a classroom simulation. Not a sandbox. By our 3rd Saturday, your team’s labels will have measurably improved the next model we deploy.
We’re looking for you because fresh eyes bring fresh insight. If you can tell two images of settling sludge apart, you already have the most important skill. Everything else — the activated sludge process, the labeling tools, how the model learns — we’ll teach you in three Saturday mornings. Your keen observation, consistency, and willingness to learn are what we’re after; good gaming skills and strategic mind will not be wasted.
Bring your laptop and make sure your WiFi and USB-A are working. You will receive a welcome envelope at the door containing a USB key with your login credentials and the class materials you’ll use during the workshop.
Sessions will be held in San Jose, California. The specific venue address will be shared with accepted applicants in the decision email.
| Time | Session | What happens |
|---|---|---|
| 9:00 – 9:10 | Welcome & icebreaker | Names, team introductions, and team color assignments. You'll be on one of four teams: Red, Blue, Green, or Yellow. |
| 9:10 – 9:50 | Wastewater primer (40 min) | How wastewater treatment works, why microbes matter, what the Settleomatic™ measures, and how AI fits into the picture. Ends with: “You are about to teach an AI.” |
| 9:50 – 10:20 | Settleomatic™ hardware demo (30 min) | Jack walks through the pre-production build (live or remote, depending on day-of logistics). See the actual hardware that captures the settling images you will be labeling. Q&A with Jack. |
| 10:20 – 10:30 | Break & snack | |
| 10:30 – 11:10 | Tools setup (40 min) | Open your welcome envelope. Install WireGuard, import your config, connect to the network. Then open Grafana and log in. Chromebooks are fine if they support USB-A. |
| 11:10 – 11:40 | Exemplary sessions tour (30 min) | The instructor walks through real settling images — one example of each classification type. You follow along on your own laptop. |
| 11:40 – 11:50 | Competition & leaderboard (10 min) | How scoring works: your labeled sessions earn points. Three awards at the end: Most Trained, Most Accurate, Best Team. |
| 11:50 – 11:58 | Your first labels (8 min) | Label two real sessions together as a group. Confirm you see the green Save confirmation. You’re now officially training the AI. |
| 11:58 – 12:00 | Wrap & homework | Receive your team’s session range. Goal: 30+ labeled sessions before May 9. |
| 12:00 – 12:30 | Lunch | Provided. |
Goal: 30+ labeled sessions per student before May 9. At ~2 min per session this is approximately 60–90 minutes of total labeling time — split comfortably across Tuesday and Thursday evenings. Use the Team filter on the dashboard to see your assigned sessions. Team Zoom check-in: Thursday May 8 at 5:00–5:30 PM.
Each team receives an identical distribution of session difficulty — normal, elevated, bulking, denitrification, and pin floc. Your team decides its own labeling strategy: focus on quantity, target the hardest cases for accuracy, or a mix of both. The leaderboard measures which strategy best improves the AI.
Bonus point multipliers by calibration group: Cal-A3 & Cal-A4 = 3× points | Cal-A1 = 2× points | Cal-A2 = 1× points. Harder calibration groups are weighted higher because the AI needs more training data in those categories.
| Time | Session | What happens |
|---|---|---|
| 9:00 – 9:10 | Welcome back | Quick round: “How was homework?” Top contributors from Week 1 get a shout-out. |
| 9:10 – 9:40 | 1st batch results (30 min) | See the before/after accuracy chart. Your labels from last week already moved the model. Which categories improved? Which still need work? |
| 9:40 – 10:30 | ML deep dive (50 min) | How your label becomes a row in the database, feeds the training script, nudges the model weights, and flows back as better predictions in Grafana. The full loop — shown live with real data. |
| 10:30 – 10:40 | Break & snack | |
| 10:40 – 10:55 | Diagnose the gaps (15 min) | Which settling conditions is the AI still getting wrong? Your team picks a target from the deficiency map. |
| 10:55 – 11:55 | Team labeling — hard cases (60 min) | One hour of focused labeling on the hardest sessions. Pair up if you want: one labels, one verifies. Live leaderboard on screen. |
| 11:55 – 12:00 | Wrap & homework | Day’s stats shown. Homework Week 2 assigned. |
| 12:00 – 12:30 | Lunch | Provided. |
Continue same team ranges from Week 1 + 5 deliberately difficult sessions per person. Zoom team huddle Thursday May 15, 5:00–5:30 PM. Estimated time: ~60–90 min total across Tue & Thu evenings.
| Time | Session | What happens |
|---|---|---|
| 9:00 – 9:15 | 2nd batch results | Three bars on one chart: before the workshop, after Week 1, after Week 2. Your two weeks of work, visible and measurable. |
| 9:15 – 9:40 | Open discussion (25 min) | What would you change about the dashboard or the labeling tool? Every idea goes on the whiteboard. The group votes on the top 2–3 to build live. |
| 9:40 – 9:50 | Break & snack | (AI working session setup while you stretch) |
| 9:50 – 10:50 | Live AI working session (60 min) | The workshop’s centerpiece. Jason and Claude build the top-voted improvement live, in front of the group. You participate — deciding what to accept, what to question, and what to change. If the AI makes a mistake, we don’t hide it. That’s the best teaching moment. |
| 10:50 – 11:00 | Reflect | Where did human judgment matter most today? What would have gone wrong if we had just trusted the AI? |
| 11:00 – 11:10 | Break | (Final setup for awards) |
| 11:10 – 11:50 | Awards ceremony | 3-min recap of what the group built together. Three awards: Most Trained, Most Accurate, Best Team. Every participant receives an AI Trainer certificate. Group photo and cake. |
| 11:50 – 12:00 | Closing & next steps | What comes next for the project and how you can stay involved. |
| 12:00 – 12:30 | Lunch & celebration | Provided. |
| Component | Days | Per session | Total |
|---|---|---|---|
| Saturday sessions (on-site) | 3 | 3 hrs (9 AM–12 PM) | 9 hrs |
| Lunch after each Saturday | 3 | 30 min | 1.5 hrs |
| Weekday labeling — Tue & Thu (remote) | 4 | ~1 hr | 4 hrs |
| Thursday Zoom check-ins (remote) | 2 | 30 min | 1 hr |
| Homework prep & weekly report | Ongoing | ~30 min/week | ~1 hr |
| TOTAL | 9 days | ~16.5 hrs |
Every participant receives an AI Trainer certificate, a portfolio of labeled sessions and AI accuracy charts, and a letter of recommendation. Top 3 winners additionally receive a ribbon, an award certificate, and a small prize.
ONK Technologies delivers advanced analytics and control for wastewater treatment plants — replacing manual sampling with continuous real-time intelligence. Our Settleomatic™ system automates the entire SV30 settling test every hour. We are currently conducting active pilot deployments at metropolitan treatment facilities and preparing for our showcase at WEFTEC-2026.
Intelligent wastewater automation — hardware, software, and AI services for municipal and industrial treatment.
R&D partner: embedded systems, ML/AI algorithms, and real-time SCADA signal processing.
Manufacturing partner: industrial systems, environmental scrubbers, and process controls.
AI Trainer certificate signed by ONK Technologies Inc. workshop leaders.
Labeled session logs and AI accuracy improvement charts for your resume.
From workshop leaders for college, graduate school, or job applications.
You will be added to WhatsApp groups on Day 1 for team messaging and coordination throughout the program. Weekly check-ins are held on Zoom — Thursdays 5:00–5:30 PM. Each team chooses its own training strategy — the goal is to discover which approach best improves AI accuracy.
| Group | Members | Purpose | Check-in |
|---|---|---|---|
| All Participants | All 10 + Jason Kim & MiSook Chung (admins) | Announcements & milestones | As needed |
| Team Red | 2–3 participants + lead | Coordinate strategy & labeling e.g. targeted training — focus on hardest cases first | Zoom · Thu 5:00–5:30 PM |
| Team Blue | 2–3 participants + lead | Coordinate strategy & labeling e.g. brute force training — maximize volume across all types | Zoom · Thu 5:00–5:30 PM |
| Team Green | 2–3 participants + lead | Coordinate strategy & labeling e.g. randomized training — diverse sampling across sessions | Zoom · Thu 5:00–5:30 PM |
| Team Yellow | 2–3 participants + lead | Coordinate strategy & labeling e.g. balanced training — equal effort per classification | Zoom · Thu 5:00–5:30 PM |
Jason designed the Settleomatic AI model. During the workshop he explains how the model interprets settling images, reviews what it gets right and wrong on your labels, and leads the Saturday #3 live improvement session where your top-voted ideas become real code changes.
Seung introduces you to the activated sludge process and shows you what healthy vs. failing tanks look like — so you can calibrate your eye for the subtle signals the model is trying to learn. His 36 years at the City of Los Angeles Bureau of Sanitation is the authority behind every image you’ll be looking at.
Jack manufactures the Settleomatic™ pre-production hardware and runs the Day 1 hardware walk-through. His hazmat training keeps the field side of the data pipeline safe — sludge samples are collected, handled, and processed at the plant before the images ever reach you, so your work happens entirely in the comfort and safety of the office.
MiSook runs your workshop experience end-to-end — schedule, meals, team check-ins, certificates, and letters of recommendation. She’s also the contact point for accommodations, parental consent forms, and anything else that needs an adult-in-charge to sort out.
As a participant you will be granted access to proprietary tools, data systems, and infrastructure. By participating, you agree to the following:
Violations will result in immediate removal from the program, revocation of all system access, and forfeiture of workshop privileges. Serious violations may be escalated to appropriate authorities and/or law enforcement.
* Participants under 18 years of age require written parent or guardian authorization prior to attendance. Authorization forms will be provided upon acceptance.
For printed version