Remnant-Loop: From Waste to Worth

🌱 The Inspiration

Every harvest leaves behind agricultural residue: paddy straw, wheat straw, sugarcane trash, and other crop by-products. Yet material that could potentially become a valuable resource is often difficult to recover efficiently because farmers, biomass buyers, transporters, and processors operate in disconnected networks.

Farmers may struggle to find suitable buyers, while biomass-based enterprises need reliable information about material type, quantity, location, and collection availability. Even when a transaction takes place, proving how much biomass was actually collected and whether it genuinely contributed to avoiding open-field burning can be difficult.

We wanted to address the problem beyond simply creating another agricultural marketplace.

What if agricultural residue could move through one connected system, from field discovery to buyer matching, verified collection, and evidence-based impact reporting?

That question inspired Remnant-Loop, an AI-powered agricultural biomass upcycling and micro-enterprise network.

🌾 What Is Remnant-Loop?

Remnant-Loop connects agricultural residue suppliers with potential buyers and supports the operational journey from listing to collection.

The platform combines AI-assisted visual inspection, explainable spatial matching, collection coordination, digital traceability, and evidence-tiered environmental accounting.

Our initial regional focus is the Karnal agricultural cluster in Haryana, India, with local supply discovery centred on areas such as Taraori, Nilokheri, Gharaunda, and Indri.

Rather than treating residue as an isolated listing, Remnant-Loop treats it as a resource moving through a traceable supply chain.

⚙️ How We Built It

1. AI-Assisted Residue Inspection

Farmers can upload photographs of agricultural residue and provide field specifications such as material type, acreage, reported quantity, storage conditions, and locality.

We integrated Google's Gemini Vision API to assist with material classification, observable quality cues, possible contamination indicators, and potential upcycling applications.

An important design decision was to distinguish visual observations from physical measurements. A photograph alone cannot establish the exact weight, dry mass, or internal moisture content of a residue pile. The system therefore preserves the quantity's provenance and does not treat an image-based guess as a verified measurement.

2. Explainable Spatial Matching

A marketplace is only useful when the right supply reaches the right buyer.

Our matching engine ranks potentially suitable biomass listings using five factors:

• Material compatibility: Does the residue match the buyer's requested material?
• Proximity: How far is the listing from the buyer or collection point?
• Quantity fit: How closely does the available quantity match the requirement?
• Pickup window: Is collection feasible within the available timeframe?
• Data completeness and freshness: How complete and current is the listing information?

The scoring breakdown helps users understand why a listing ranks highly instead of relying on an unexplained recommendation.

3. Regional Supply Discovery

The Regional Map provides a geographical view of biomass listings across the pilot cluster.

Users can explore supply locations and distinguish available, reserved, and collected lots. This spatial view supports local sourcing and creates a foundation for future multi-farm aggregation and coordinated transport planning.

4. Logistics and Dual-Confirmation Collection

A reservation is not the same as a completed collection.

Remnant-Loop includes a logistics workflow for creating dispatch jobs, scheduling pickups, and recording collection evidence. Final collection status requires both farmer handover confirmation and buyer or collector weight evidence.

The system also supports safeguards against duplicate reservations and duplicate weighbridge ticket references, helping improve transaction integrity and auditability.

5. Residue-to-Resource Passport

Our signature feature is the Residue-to-Resource Passport, which brings the lifecycle of a biomass lot into one structured record.

The passport is designed to connect:

Farm origin → AI-assisted inspection → Buyer matching → Reservation → Collection confirmation → Weight evidence → Eligible impact reporting.

This creates a clearer history of what was listed, what was reserved, what was collected, and which evidence supports each stage.

6. Evidence-Tiered Environmental Impact Ledger

Environmental impact should be supported by evidence, not just attractive dashboard numbers.

Remnant-Loop distinguishes simulated records, farmer-reported quantities, transaction-confirmed collections, and independently measured quantities. Demo records are kept separate from real-world impact totals, and environmental calculations are restricted to records that satisfy the configured evidence requirements.

The accounting methodology must also distinguish emissions avoided under a stated baseline from emissions actually measured, and account for relevant transport emissions and end-use assumptions.

The goal is to make impact reporting transparent, explainable, and open to scrutiny.

🧠 Technology Stack

We built Remnant-Loop as a full-stack application with a modular backend and a dashboard-oriented frontend.

• Frontend: React, TypeScript, Vite
• Backend: Python, FastAPI
• AI and computer vision: Google Gemini Vision API
• Database: PostgreSQL / Tiger Data, with SQLite support for local development and testing
• Geospatial matching: Coordinate-based proximity calculations and explainable multi-factor ranking
• Persistent preferences: Backboard integration for consent-based user preferences and memory
• Deployment: Docker-based deployment configuration, with Vultr as the target hosting platform
• Testing: Automated backend tests covering listings, matching, reservation concurrency, collection confirmation, impact summaries, and the residue passport

The system is designed to keep the business logic, AI inspection, matching engine, and evidence records as separate components, making the platform easier to test and extend.

🚧 Challenges We Faced

Reliable quantity estimation

One of the biggest challenges was deciding what AI should and should not infer from an image. Visual appearance varies with lighting, camera angle, pile density, and moisture. We therefore avoided treating photographs as a reliable source of exact tonnage.

Matching fragmented supply with real demand

Small agricultural lots may not individually satisfy a larger buyer's procurement requirement. We designed the matching workflow around material compatibility, proximity, quantity fit, collection timing, and data quality, creating a foundation for future multi-farm pooling.

Preventing false collection claims

A listing or reservation does not prove that material reached its destination. The dual-confirmation workflow and evidence records help distinguish intended transactions from completed collections.

Making environmental claims defensible

Calculating avoided emissions requires more than multiplying listed tonnes by a fixed factor. The baseline, dry-matter basis, relevant gases, transport emissions, end-use pathway, and evidence quality all matter. We designed the impact ledger to preserve these distinctions rather than automatically converting every listing into a sustainability claim.

Integrating multiple services

Combining an external vision API, a structured operational database, persistent user preferences, and a multi-stage logistics workflow required careful handling of service failures, data consistency, and fallback behaviour.

📚 What We Learned

Building Remnant-Loop reinforced several important lessons:

• AI is most useful when its role and limitations are clearly defined.
• Explainable matching can make recommendations easier to understand and challenge.
• Operational verification is essential when a product makes claims about physical-world activity.
• Environmental accounting needs a documented methodology and appropriate evidence.
• A successful sustainability platform must connect the entire journey, from supply discovery to actual recovery, rather than stopping at awareness or listing creation.

We also learned that a credible environmental technology product must be designed around uncertainty, provenance, and accountability from the beginning.

🚀 What's Next?

Our next priorities are to expand the labelled agricultural residue image benchmark, evaluate vision outputs against human-confirmed labels, validate environmental accounting assumptions, test integrations in the deployed environment, and obtain real-world feedback from farmers, aggregators, and biomass buyers.

Future development opportunities include multi-farm pooling, improved collection-route planning, richer residue traceability, and expansion to additional agricultural clusters.

🌍 Our Vision

We envision a connected ecosystem in which agricultural residue becomes easier to discover, evaluate, exchange, and trace.

Remnant-Loop is built around one principle: turn agricultural residue into a potential resource, connect the people who can use it, and report outcomes only as strongly as the evidence allows.

From field residue to real impact.

Built With

• agriculture
• artificial-intelligence
• cloud-computing
• computer-vision
• data-visualization
• docker
• environmental-monitoring
• fastapi
• geospatial-analysis
• google-gemini-api
• postgresql
• python
• react
• rest-api
• supply-chain
• sustainability
• tiger-data
• typescript
• vite