Rivet

Inspiration

What if a robot could save both lives and livelihoods? The International Labour Organization estimates 2.93 million annual deaths globally from occupational injuries and diseases. Workers face hazards ranging from high-altitude inspections to toxic waste handling. In California, a worker drowned while clearing weeds from a golf course pond. Robotics is already demonstrating its potential: SludgeHammer's PondBot reportedly cleared algae from a 30-acre pond in three weeks, while Marsden Services reported $12,000 in monthly labor savings and an 18-month payback period using autonomous cleaning robots.

Blue-collar industries offer clear opportunities for automation. Already, 77% of manufacturers surveyed by the Society of Manufacturing Engineers recognize automation's value, yet the National Institute of Standards and Technology identifies cost and integration barriers to adoption. Meanwhile, robotics startups face a commercialization valley of death, struggling with product-market fit, financing, and uncertain demand, as highlighted in the National Security Commission on Robotics for Advanced Manufacturing's 2026 report.

Why not "rivet" these two worlds together? Group purchasing and competitive vendor bids could lower adoption costs, while shared development bounties could fund unmet needs and validate demand for robotics companies. Rivet aims to empower grassroots blue-collar businesses and emerging robotics companies while helping workers transition into safer technical roles operating and maintaining robots, ultimately benefiting local communities.

Features

Our platform connects small blue-collar businesses in infrastructure maintenance, property services, construction, and other labor-intensive industries with robotics startups and engineers worldwide. Rivet reimagines robotics shopping through agentic procurement, engineering-based matching, and a bounty marketplace, connecting real operational needs with existing products or new development opportunities.

We have three main functionalities: AI-powered robotics discovery, demand-driven bounties, and developer matching. For robotics discovery, buyers can describe operational challenges to our AI agent or use manual search filters. They can also upload videos of daily workflows and Excel financial records, including labor costs, operating expenses, and assets. Our agent decomposes operational tasks into engineering requirements, performs constraint-aware matching, and ranks existing robotics solutions by technical feasibility, deployment readiness, and economic viability. Buyers can compare vendor quotations, total cost of ownership (TCO), projected return on investment (ROI), and payback periods.

For the bounty marketplace, when no suitable or affordable solution exists, our agent identifies unmet engineering requirements and asks targeted follow-up questions to generate structured bounties with technical specifications, deliverables, and funding targets. Compatible requests can be aggregated into shared bounties, pooling potential funding and validating market demand. Businesses can also post unfunded requests to attract developer interest before securing capital.

For developer matching, robotics startups and independent engineers can list existing products, receive AI-matched customer requests, submit engineering proposals, and claim development bounties. Integrated messaging enables direct technical discussions, quotations, and project coordination. By connecting buyers and developers globally, Rivet expands access to diverse robotics technologies, engineering expertise, and competitive pricing across markets, creating a continuous pathway from operational demand to robotics deployment and commercialization.

How we built it

We built Rivet on Convex for real-time backend queries, data synchronization, and buyer-developer messaging. We integrated the Claude API for workflow analysis, Gemini API for video processing, and Apify for live robotics product data. Gemini analyzes workplace footage to identify equipment and track worker movements. It also estimates spatial relationships and detects repetitive or hazardous tasks. Claude combines Gemini's video analysis with user voice or text descriptions and uploaded Excel financial records covering labor costs, revenue, operating expenses, and assets. It then maps the business's operations into a step-by-step workflow graph. Each step captures its task dependencies, labor requirements, operating costs, and safety risks. Each step is then evaluated for robotic automation feasibility and projected ROI. This includes potential labor savings, deployment costs, and payback periods. Our recommendation engine matches these opportunities with robotics products retrieved through Apify. It combines hybrid semantic search with hard-constraint filtering to exclude products that fail payload, terrain, operating environment, or budget requirements. Eligible robots are ranked by task compatibility, deployment readiness, and economic viability.

We use three mathematical models to evaluate robotics solutions, calculate financial returns, and aggregate unmet demand.

For constraint-aware solution matching, we rank robots using hard-constraint feasibility (Fⱼ), capability match (Cⱼ), deployment readiness (Rⱼ), and economic attractiveness (Eⱼ), with initial heuristic weights: Sⱼ = Fⱼ(0.4Cⱼ + 0.3Rⱼ + 0.3Eⱼ).

For financial evaluation, we calculate total cost of ownership (TCO), return on investment (ROI), and net present value (NPV) using discounted cash flow: NPV = −C₀ + Σₜ₌₁ᴺ [(Bₜ − Oₜ)/(1 + r)ᵗ], where Bₜ represents annual benefits, Oₜ operating costs, C₀ upfront investment, and r the discount rate.

For shared bounty aggregation, we cluster technically compatible buyer requests (Cₖ) and sum their proposed contributions (bᵢ) using Bₖ = Σ(i ∈ Cₖ) bᵢ to estimate pooled development funding while distinguishing proposed contributions from committed capital.

Challenges we ran into

Our biggest challenge was reliable data processing across multiple AI agents and backend services. Company Excel spreadsheets contained inconsistent formats, missing values, and formula errors that distorted ROI calculations. We implemented a multi-agent validation pipeline to detect anomalies, flag uncertain values, and cross-check financial calculations through targeted chatbot follow-ups. Another challenge was designing reliable handoffs between search and reasoning layers. Each search needed to distinguish no matching robot exists from search execution failed. The first triggers bounty generation, while the second triggers retries or fallback retrieval. We separated the pipeline into fast retrieval, data processing, deterministic calculations, and deep reasoning to reduce latency, API costs, and numerical hallucinations. Finally, coordinating Convex backend operations with external AI APIs required managing asynchronous requests, inconsistent response formats, and state synchronization to prevent incomplete results from disrupting downstream workflows.

Accomplishments that we're proud of

We're most proud of our multimodal video analysis pipeline, which combines object detection, motion tracking, spatial reasoning, and workflow decomposition to turn raw workplace footage into actionable robotics requirements, eliminating the need for small businesses to manually define complex engineering specifications they may lack the expertise to articulate.

What's next for Rivet

Going forward, we plan to onboard robotics companies and negotiate platform-exclusive discounts for buyers. We also aim to integrate manufacturers so engineers can compare production costs, completing the pipeline from blue-collar businesses to engineers, engineers to manufacturers, and finished products back to businesses. Our business model can focus on success-based referral fees from completed transactions.