Read-only tools giving the agent live access to each user’s account
Guided playbooks, for event producers and first-time speakers
Secrets or user emails sent to the model
Edit on the Spot
Claymont, Delaware
Event Technology / Live Video Production
AWS Bedrock (gpt-oss-120b via OpenAI-compatible endpoint), MCP-style tool contracts, tRPC, TypeScript, React / Next.js, PostgreSQL / Prisma, AWS Amplify (test environment)
Edit on the Spot operates Eventcut, a cloud-based platform that captures multi-room live streams and turns them into branded, segmented video in near real time. The platform’s depth, real-time ingest, AI transcription, automatic segment detection, and one-click export, meant new users often could not reach their first successful outcome without direct human help, creating friction in both onboarding and sales. Avahi built an AI-powered onboarding copilot directly inside the Eventcut web application: a persona-aware conversational agent on AWS Bedrock, backed by a read-only tool layer with MCP-style contracts, designed to become a standalone MCP server, with live access to a user’s account state, paired with a UI “spotlighting” overlay that visually highlights the exact element a user needs to click next. The work was delivered as a self-contained, PR-ready branch for Edit on the Spot’s engineering team to review and merge on their own timeline.
Edit on the Spot is an AI video infrastructure company. Its platform, Eventcut, captures multi-room live streams and delivers branded, segmented video in near real time, serving event producers, A/V teams, marketing managers, and speakers.
Eventcut’s feature depth, real-time RTMP/SRT ingest, AI transcription, multi-modal segment detection, smart topic clipping, brand kit management, meant new users frequently struggled to reach their first successful outcome without direct human assistance. Two moments created the highest-stakes friction: an Event Producer setting up a live stream shortly before an event needed to create a stream key and configure their encoder correctly the first time, and a Speaker arriving via an emailed session link, often with no prior platform experience, needed to find their session and review their first video without help. Without a scalable way to guide these two groups through their first task, the platform’s sales and onboarding motion depended on direct human support.
Eventcut’s existing application already integrated with an OpenAI-compatible SDK, so routing the new onboarding agent through AWS Bedrock’s OpenAI-compatible endpoint required only a base URL change rather than a new integration layer. This let the agent run on AWS infrastructure from day one while reusing the platform’s existing request and authentication patterns.
Edit on the Spot needed the new onboarding capability built as a natural extension of its existing product, not a bolt-on tool that introduced a second system to maintain. Avahi designed the integration to wrap Eventcut’s existing tRPC procedures directly rather than standing up a separate API, and to deliver the copilot as a separate chat panel that replaces the existing support widget. That minimal-footprint approach, combined with Avahi’s experience building production-grade agents on AWS Bedrock, gave Edit on the Spot a path to a working onboarding agent without re-architecting their application.
Avahi built a read-only tool layer, with six tools using MCP-style contracts, that wraps Eventcut’s existing tRPC procedures and runs inside the application, giving the onboarding agent live access to a user’s stream keys, projects, team roles, and setup progress, so its guidance reflects the user’s actual account state rather than generic scripted steps. The agent runs on AWS Bedrock via an OpenAI-compatible endpoint and determines whether a user is an Event Producer or a Speaker, first from account signals available through the tool layer, falling back to a single clarifying question only when that signal is ambiguous.
The agent responds with a structured message plus a spotlight target drawn from a closed set of real, tagged UI elements, so it can only ever point users at controls that actually exist on screen. A lightweight attribute added to the Eventcut frontend lets a spotlight component resolve that target, dim the rest of the screen, and draw a highlight ring around it, falling back gracefully to a text-only instruction if the target element isn’t present. The agent’s FAQ knowledge base is grounded in content pulled from a client-designated source document, so answers stay scoped to Edit on the Spot’s own guidance rather than general or invented information.
The design limits what the model can see. No secrets or user email addresses are sent to the model, and every tool call runs with the signed-in user’s own permissions, so the copilot can only access what that user could already access.
Read-only tool layer with six MCP-style tools for live account state (user profile, org setup state, stream keys, projects, team members, integrations), designed to become a standalone MCP server
Persona-aware onboarding agent on AWS Bedrock with conversation and persona-detection logic
UI spotlighting overlay component, delivered within the new copilot chat panel that replaces Eventcut’s existing support widget
Two end-to-end guided playbooks (Event Producer stream setup; Speaker first-clip creation) integrated and validated together
On-demand FAQ content refresh pipeline grounding agent answers in client-designated source material
Self-contained, PR-ready branch with inline documentation for independent review and merge
Recorded knowledge-transfer session covering agent architecture, playbook authoring, and tool layer extension
Avahi delivered a working, internally validated onboarding copilot covering both in-scope personas, packaged as a clean branch Edit on the Spot’s engineering team can review and merge independently. The read-only tool layer was built as a reusable pattern: because it uses MCP-style contracts, it is designed to become a standalone MCP server and to extend with write-capable tools and additional workflows in future work, without changing how the agent connects to the platform.
Edit on the Spot
Claymont, Delaware
Event Technology / Live Video Production
AWS Bedrock (gpt-oss-120b via OpenAI-compatible endpoint), MCP-style tool contracts, tRPC, TypeScript, React / Next.js, PostgreSQL / Prisma, AWS Amplify (test environment)
Edit on the Spot operates Eventcut, a cloud-based platform that captures multi-room live streams and turns them into branded, segmented video in near real time. The platform’s depth, real-time ingest, AI transcription, automatic segment detection, and one-click export, meant new users often could not reach their first successful outcome without direct human help, creating friction in both onboarding and sales. Avahi built an AI-powered onboarding copilot directly inside the Eventcut web application: a persona-aware conversational agent on AWS Bedrock, backed by a read-only tool layer with MCP-style contracts, designed to become a standalone MCP server, with live access to a user’s account state, paired with a UI “spotlighting” overlay that visually highlights the exact element a user needs to click next. The work was delivered as a self-contained, PR-ready branch for Edit on the Spot’s engineering team to review and merge on their own timeline.
Edit on the Spot is an AI video infrastructure company. Its platform, Eventcut, captures multi-room live streams and delivers branded, segmented video in near real time, serving event producers, A/V teams, marketing managers, and speakers.
Eventcut’s feature depth, real-time RTMP/SRT ingest, AI transcription, multi-modal segment detection, smart topic clipping, brand kit management, meant new users frequently struggled to reach their first successful outcome without direct human assistance. Two moments created the highest-stakes friction: an Event Producer setting up a live stream shortly before an event needed to create a stream key and configure their encoder correctly the first time, and a Speaker arriving via an emailed session link, often with no prior platform experience, needed to find their session and review their first video without help. Without a scalable way to guide these two groups through their first task, the platform’s sales and onboarding motion depended on direct human support.
Eventcut’s existing application already integrated with an OpenAI-compatible SDK, so routing the new onboarding agent through AWS Bedrock’s OpenAI-compatible endpoint required only a base URL change rather than a new integration layer. This let the agent run on AWS infrastructure from day one while reusing the platform’s existing request and authentication patterns.
Edit on the Spot needed the new onboarding capability built as a natural extension of its existing product, not a bolt-on tool that introduced a second system to maintain. Avahi designed the integration to wrap Eventcut’s existing tRPC procedures directly rather than standing up a separate API, and to deliver the copilot as a separate chat panel that replaces the existing support widget. That minimal-footprint approach, combined with Avahi’s experience building production-grade agents on AWS Bedrock, gave Edit on the Spot a path to a working onboarding agent without re-architecting their application.
Avahi built a read-only tool layer, with six tools using MCP-style contracts, that wraps Eventcut’s existing tRPC procedures and runs inside the application, giving the onboarding agent live access to a user’s stream keys, projects, team roles, and setup progress, so its guidance reflects the user’s actual account state rather than generic scripted steps. The agent runs on AWS Bedrock via an OpenAI-compatible endpoint and determines whether a user is an Event Producer or a Speaker, first from account signals available through the tool layer, falling back to a single clarifying question only when that signal is ambiguous.
The agent responds with a structured message plus a spotlight target drawn from a closed set of real, tagged UI elements, so it can only ever point users at controls that actually exist on screen. A lightweight attribute added to the Eventcut frontend lets a spotlight component resolve that target, dim the rest of the screen, and draw a highlight ring around it, falling back gracefully to a text-only instruction if the target element isn’t present. The agent’s FAQ knowledge base is grounded in content pulled from a client-designated source document, so answers stay scoped to Edit on the Spot’s own guidance rather than general or invented information.
The design limits what the model can see. No secrets or user email addresses are sent to the model, and every tool call runs with the signed-in user’s own permissions, so the copilot can only access what that user could already access.
Read-only tool layer with six MCP-style tools for live account state (user profile, org setup state, stream keys, projects, team members, integrations), designed to become a standalone MCP server
Persona-aware onboarding agent on AWS Bedrock with conversation and persona-detection logic
UI spotlighting overlay component, delivered within the new copilot chat panel that replaces Eventcut’s existing support widget
Two end-to-end guided playbooks (Event Producer stream setup; Speaker first-clip creation) integrated and validated together
On-demand FAQ content refresh pipeline grounding agent answers in client-designated source material
Self-contained, PR-ready branch with inline documentation for independent review and merge
Recorded knowledge-transfer session covering agent architecture, playbook authoring, and tool layer extension
Avahi delivered a working, internally validated onboarding copilot covering both in-scope personas, packaged as a clean branch Edit on the Spot’s engineering team can review and merge independently. The read-only tool layer was built as a reusable pattern: because it uses MCP-style contracts, it is designed to become a standalone MCP server and to extend with write-capable tools and additional workflows in future work, without changing how the agent connects to the platform.
Let’s explore your high-impact AI opportunities together in a complimentary session
