ML · Forecast
Predicts when a bin will overflow.
A gradient-boosted ensemble — fastest model class in its tier for tabular time-series. Retrains in minutes.
1-hour MAE ±4.13%
6-hour MAE ±9.83%
R² (1h) 0.73
Issue 01 · Cebu City, 2026 Multi-tenant IoT · Field report
Every bin, a signal.
EcoRoute is a smart-bin network for subdivision-level waste management — three ultrasonic sensors per bin, a multi-tenant dashboard for HOA admins, and collection routes that plan themselves around the trash that's actually there.
Filed from
Barangay Buhisan
Cebu City · 10.299° N
Bins tracked
0
Subdivisions
0
Readings / day
0
Avg route distance
0.0 km
Fixed schedules,
variable reality.
Most subdivisions in Cebu City still run waste collection on a fixed weekly calendar — Tuesdays and Fridays, rain or shine, full or empty. It's a system that served the 1990s, when the alternative was nothing. It does not serve 2026, when the city generates an estimated 700 tonnes of solid waste per day.
Some bins overflow by Thursday. Others sit half-empty on pickup day, burning diesel and crew time for nothing. There is no feedback loop — trucks leave the depot without knowing what they're collecting, and come back without a record of what they actually moved.
The cost falls unevenly. Residents see flies and smell garbage near missed pickups. Barangay officers spend their Saturdays fielding complaints that were preventable on Wednesday. And the hauliers, legally bound under RA 9003, run routes whose optimization stopped improving a decade ago.
The premise
What if every bin
told us, in real time,
how full it actually was?
The question that started EcoRoute
Ch. 02
Sense. Predict. Route.
How it works.
STEP 01
Sense the surface.
Inside each lid, three ultrasonic transducers form a triangle. They fire in sequence — one, then another, then the third — so their echoes don't collide. Each one measures the distance to the trash beneath it independently.
A single sensor sees one spot. Three sensors see a surface. That's how we catch an uneven pile at the back of the bin that a center-mounted sensor would miss until the trash tipped over.
Fig. 01 — Triangular sensor layout, lid cross-section
Fig. 02 — Historical fill vs. forecast curve
STEP 02
Predict the fill.
Every bin's history is a sawtooth — slow rise through the week, sharp drop at collection. That pattern is legible. Given today's curve and 30 days of prior slopes, a small model estimates when each bin will cross its threshold next.
Dispatchers see it as a countdown: Bin AHB-003 will overflow in 11 hours. The right truck can be on its way before anyone complains.
STEP 03
Route the truck.
Only the bins that actually need servicing make it onto tomorrow's route. EcoRoute feeds the shortlist into OpenRouteService and returns a solved sequence — depot, stops in order, return — with distance, duration, and optimization score.
The driver sees the route on a phone. Photo-evidence and timestamps of each pickup close the loop back to the dashboard, so the next prediction is a little better than the last.
Fig. 03 — Optimized collection route
Ch. 03
Six-hundred-peso engineering, open-source all the way down.
The bin itself.
Fig. 04 — Bin cross-section, lid partially open
01
Triple ultrasonic array
Three RCWL-9610A transducers, triangulated on the lid underside, fired in sequence with a 120 ms settling delay.
02
HW-201 IR lid sensor
Gates the ultrasonic reads. If the lid is open, we don't trust the signal — we hold the last stable value.
03
Wire harness
Seven signal wires through the hinge — three TRIG, three ECHO, one IR — plus shared 3V3 and ground.
04
ESP32-WROOM
In a sealed compartment at the base. Publishes to MQTT every 1–10 min depending on fill trajectory.
05
The actual trash
Never evenly distributed. The whole reason three sensors beat one — you catch the piles a single center-point sensor would miss.
Firmware
routeOS v3.2.0-tc3
Accuracy
~90% (3-sensor fusion)
Telemetry
MQTT · JSON
Per-bin BoM
≈ ₱1,200
Power
Battery + solar
Ch. 04
A dashboard that shows where the trash is, not just how much.
The dashboard.
Multi-tenant. Scoped per subdivision. And — because one point can't tell you where trash is piled — visualized as a live heatmap, interpolated from the three sensors in every bin.
ecoroute.yeems214.xyz / bins / ECO-BIN-AHB-003
LIVE
BIN ID
ECO-BIN-AHB-003
10.2991° N, 123.8645° E
Anika Homes Buhisan
Fill level
—
Battery
3.72V
Signal
−68dBm
Last reading
just now
Fill Distribution Heatmap
Interpolated from 3 ultrasonic sensors · IDW, power 4
empty half high full
INTERPOLATED
—
Move your cursor over the heatmap to probe interpolated values. Tap and hold the heatmap to probe interpolated values.
Per-sensor
Activity
Other bins in Anika Homes Buhisan
Every value above is simulated in your browser. The same component, wired to real MQTT telemetry, runs at ecoroute.yeems214.xyz.
Ch. 05
The intelligence layer — ML plus a language model on top.
AI / Machine Learning
Built on the fastest,
state-of-the-art
models.
EcoRoute runs two intelligence services side-by-side: a fast, ensemble machine-learning pipeline that predicts when each bin will fill, and a large language model that translates live telemetry into plain-language summaries for HOA admins who have never read a dashboard before.
ML · Classify
Flags the bins you must collect today.
Actual
Predicted
TP 0 caught
FN 0 missed
FP 0 detour
TN 0 skipped
Tuned for recall over precision — missing a full bin is more expensive than one extra detour.
Accuracy 93%
Recall 80%
Horizon 6 hours
AI · Explain
Turns telemetry into plain language.
Insights generated this week 0
Mon Tue Wed Thu Fri Sat Sun
AI-GENERATED · today
"Three bins in the pilot cluster are approaching threshold in the next 4 hours — all near the south entrance. Route truck 02 there before 11 AM to avoid the Friday complaint window."
Powered by a large language model that reads live telemetry and writes summaries, anomaly explanations, and natural-language answers.
Model interpretability
What the forecast actually listens to.
Feature importance from the trained forecast model. The top ten out of twenty engineered inputs, normalized so the leading feature reads as 1.00.
Fill-velocity over the past hour carries the most signal — bins that are filling fast now will likely overflow fastest. Time-of-day and day-of-week encodings matter more than absolute fill level, because disposal rhythms follow household routines, not bin geometry.
Training data
8,496rows
Synthetic pre-pilot; retrained on live telemetry every quarter.
Retraining time
<5min
Full pipeline, hot-swappable with zero downtime.
Inference latency
~40ms
Per-bin prediction, on shared infrastructure.
Model family
state-of-the-art
Current benchmark leader for this class of tabular time-series work.
Current models are trained on a synthetic dataset calibrated against Department of Environment and Natural Resources residential-waste generation rates. Phase 2 scope includes full retraining on three months of pilot telemetry, at which point the accuracy numbers above become production figures rather than bench figures.
Ch. 06
It started at home.
Where the idea lives.
Fig. 05 — Target pilot area, south-west Cebu City. The orange pulse marks the neighborhood where the project began.
EcoRoute started as a neighborhood observation: the bins outside a teammate's home would overflow two nights before pickup day. Sometimes a day late. Never on time. The schedule was stable; the trash was not.
“Waste collection in our neighborhood happened on a fixed schedule that rarely matched when bins were actually full. That mismatch — that's the problem worth solving.”
The first prototype was three ultrasonic sensors glued under a pedal-bin lid, an ESP32 stuffed into the base, and a laptop running a Postgres instance on a dormitory desk. The dashboard followed. The route solver followed that.
The platform is now multi-tenant — built from day one to serve any Cebu subdivision with an HOA willing to put a bin out. One neighborhood is a pilot. A city is the point.
Ch. 07
The law was ready. The tooling wasn’t.
Aligned with Cebu policy.
2001 · NATIONAL
Republic Act 9003
Ecological Solid Waste Management Act
The national framework mandating waste segregation at source, barangay-level material recovery, and systematic planning. EcoRoute operationalizes the data loop the law already requires.
2004 · CITY
CCSWMB
Cebu City Solid Waste Mgmt. Board
Established under City Ordinance No. 2017, the board oversees policy, plans, and coordination with barangay-level committees. Our multi-tenant model mirrors this jurisdictional structure by design.
2023–2032 · ACTIVE
Cebu SWM Plan
Ten-year operational blueprint
The city’s current solid-waste plan emphasizes decision-making at the barangay level and “localized strategies.” Sensor-driven routing is exactly the kind of localization the plan asks for, without saying those words.
The policy has been in place for two decades. What’s new isn’t the intent to manage waste intelligently — it’s the hardware cheap enough to make it possible. A sensor that costs less than a merienda can now close the loop the ordinance drew.
Ch. 08
Three tiers, one architecture underneath.
The business.
Pricing that scales with real connectivity conditions — because the infrastructure between a bin in central Cebu and a bin in a mountain barangay is wildly different. Same software, same sensors, different deployment mode.
Deployment Mode 1
Cloud
₱200 / bin / month
+ ₱2,500 per bin, one-time
For subdivisions with reliable WiFi.
Best for
Urban subdivisions with existing broadband, 20–50 bins, modest HOA budget.
- Triple-ultrasonic sensing + IDW heatmap
- Multi-tenant cloud dashboard
- Real-time MQTT telemetry + alerts
- 30-min offline heartbeat buffer
- Driver mobile app with proof-of-service
- ML fill-forecast + collection classifier
- AI-generated daily summaries
- On-premises data sovereignty
- LTE cellular fallback
- Managed WiFi infrastructure
Known limits: Requires existing subdivision WiFi · shared cloud infrastructure
Most popular
Deployment Mode 2
Hybrid
₱350 / bin / month
+ ₱4,500 per bin, one-time
For operations that want data at home.
Best for
Established HOAs, 50–100 bins, data-sovereignty concerns, spotty public internet.
- Everything in Cloud
- On-premises server (Raspberry Pi 5, included)
- Data stays within subdivision network
- Full ML + analytics on-premises
- AI-generated daily summaries
- Offline operation during internet outages
- Custom alert rules + webhook integrations
- Monthly ops review with your HOA
- LTE cellular fallback
- Managed WiFi infrastructure
Known limits: Still requires subdivision WiFi for bin connectivity
Deployment Mode 3
Turnkey
Custom
+ Site-specific quote
We provide everything.
Best for
New developments, sites without WiFi, LGU-scale procurement, 100+ bins.
- Everything in Hybrid
- Managed WiFi access points at every bin
- LTE cellular backup (Phase 2) Phase 2
- IP65-certified outdoor enclosures Phase 2
- TLS-encrypted MQTT channel Phase 2
- Tamper detection (GPS + accelerometer) Phase 2
- Dedicated account manager
- 24-hour field-maintenance SLA
- Unlimited bins
- Formal privacy impact assessment Phase 2
Market context
What the rest of the world charges for a smart bin.
Public retail pricing for comparable sensor-equipped bins, converted to PHP. Our target price is one-time hardware + install on the Cloud tier.
EcoRoute Cloud tier · Cebu, PH
₱5,500 / bin
HW+INSTALL
Sensoneo Slovakia · sensors only
~₱17,000 / bin
Enevo Finland · bin sensors + SaaS
₱28–84K / unit
Bigbelly USA · compacting smart bin
~₱224K / bin
₱0 ₱100k ₱224k
International smart-bin vendors are priced for US and European procurement budgets. A Cebu subdivision pays roughly 4× to 40× less with EcoRoute for a comparable sensor footprint — plus local peso pricing, local support, and a dashboard in English rather than a language the HOA accountant has to Google-translate.
Unit economics
What a 50-bin subdivision looks like, annualized.
50 bins × ₱200 / bin / mo (Cloud tier) ₱10,000 / mo
Annual recurring revenue ₱120,000 / yr
One-time install (50 × ₱2,500) ₱125,000
Per-bin BoM cost ≈ ₱3,000
Projected 4-year LTV, per subdivision ₱480,000+
Comparable international products (Enevo, Bigbelly, Sensoneo) retail at US$500–US$1,500 per bin. EcoRoute targets a peso-priced, locally-supported alternative for the Philippine subdivision market.
For HOA boards
The first 10 subdivisions in Cebu get pilot pricing.
Free installation assessment, 60-day trial on the Cloud tier, no commitment until you see your first overflow alert resolved before anyone complains.
Try the dashboardCh. 09
Four people. One bin. A lot of coffee.
Who built this.
R
№ 01
Portrait · placeholder
Roel
Hardware Engineer
The one who actually made trash measure itself. Owns the ESP32 firmware, the triple-ultrasonic fusion, and every millimeter of the wire harness inside the lid.
ESP32 · RCWL-9610A · C++ · MQTT · PlatformIO
V
№ 02
Portrait · placeholder
Vince
AI / Backend Developer
Built the multi-tenant backend, the admin dashboard, and the IDW heatmap component. Cares about making IoT telemetry legible to people who have never touched a soldering iron.
Bun · Hono · PostgreSQL · Drizzle · React
S
№ 03
Portrait · placeholder
Seth
Mobile Developer
Shipped the Android app for drivers and dispatchers. Makes sure a truck operator with greasy hands can still close a route stop on the third tap.
Kotlin · Jetpack Compose · Hilt · Retrofit
B
№ 04
Portrait · placeholder
Bert
Project Manager
Turned three engineers into a roadmap. Keeps the feature list honest, the pilot conversations moving, and the three other names on this page pointing in the same direction.
Roadmap · Discovery · Pilot ops · Customer dev
Built in
14 weeks
Lines of code
11,200+
Open-source deps
42
Cups of coffee
uncountable