Thomas Grivet

streamk7 is the simplest and most powerful solution to increase streamer visibility across all platforms.

Using a large amount of engagement signals and multimodal data, streamk7 understands the most impactful moments, suggests ready-to-publish clips, and provides a creative space for users focused on growing their community and highlighting their content.

I built this platform after understanding the challenges and issues faced by small and medium creators who cannot afford to hire editors or spend hours after their streams sorting, filtering, and editing their videos for multiple formats and platforms.

The Front-end, API, and rendering engine are connected within a Typescript monorepo.

React with Tanstack Start, Tailwind, react-hook-form and Zod, Fastify, OpenAPI, Prisma

Render engine with mediabunny, GSAP, PixiJS, and WebGPU for the smoothest and fastest editing and viewing experience.

Rendering available client-side in the browser and server-side in on-demand GPU instances.

The highlight detection part is done in Python, running on GPU instances orchestrated via Kubernetes.

htdemucs to separate audio tracks (vocals, music), Whisper models for transcription, and NeMo for diarization.

InsightFace for emotion detection, best face detection, and vector matching.

Upscale and background removal on the best faces to reuse them in thumbnail creation.

Analysis of chat velocity and manual clips made by viewers.

Generation of embeddings for the entire stream to maximize clip relevance and offer semantic search.

Monitoring provided by Prefect.

Unit tests via Vitest and end-to-end tests with Playwright.

CI/CD with GitHub Actions.

Cocolis is a French SME specializing in collaborative parcel delivery between individuals, reducing the carbon footprint of transport (25kg of CO2 per delivery).

I was the first developer hired to lead the complete redesign of the mobile application in React Native / Expo. I designed the complete mobile architecture and laid the technical foundations before being joined by 3 colleagues. I then contributed to other platform projects: Next.js website, Shopify application, and MCP server.

Rated 4.9/5 with 12,000 user reviews, 99.8% crash-free sessions, perfect external code audits. The platform has helped avoid 12,000,000 kg of CO2.

Development of the iOS/Android app with Expo and EAS. Features: background geolocation, interactive maps (react-native-maps), multi-step forms (react-hook-form + Zod), advanced animations (Reanimated + Gesture Handler), Stripe payments, Elasticsearch, custom design system (Restyle), state management (Zustand), i18next, analytics/push (Firebase), WebSockets, Redis caching.

AI Integration: visual object recognition with API cost optimization, prompt structuring, and response schemas.

Quality: Jest/Maestro tests, MSW mocks, complete GitHub Actions CI/CD, Sentry monitoring, strict ESLint/Prettier.

App Router, Prismic CMS, Storybook, @radix-ui, advanced SEO optimizations, chatbot with Vercel AI SDK connected to the MCP server, AI object recognition, Stripe, next-intl, Jest/Playwright tests.

TypeScript server (Model Context Protocol) enabling the integration of Cocolis features into external AI tools (Claude, ChatGPT).

Remix + GraphQL app to integrate Cocolis delivery into e-commerce stores.

Yarn Workspaces Monorepo with code sharing between projects. 100% TypeScript stack, systematic testing, Linear tickets.

Development of a platform allowing users to join events organized by associations.

Used React, Typescript, and Firebase services (Firestore, Storage, Auth, Cloud Functions).

Used third-party services like Mapbox & Algolia.