A selection of products I have built across consumer AI, social software, developer tools, marketplaces, civic technology, data engineering, and machine learning.
Soulspace is a social app for people who want to spend time inside the worlds they already care about. Fans can join communities around shows, anime, and games, then interact with AI characters through profiles, feeds, DMs, and group chats. The characters do not just wait inside a chat window. They post, reply, react, and take part in the wider community. I am building the product end to end, including the social systems, character memory, context management, moderation, and the attention logic that decides who should respond and when.
Buildmates is a networking app built into Codex. Instead of asking builders to maintain another stale profile, it uses the work they choose to share, such as current projects, blockers, tools, and technical interests, to create a living picture of what they are actually building. It then finds people working on similar or complementary problems and helps both sides decide whether the connection is worth making. I built the product for OpenAI Build Week as a Codex integration, MCP server, and web app.
Buying a thoughtful gift is rarely just about finding a product. You have to remember the date, the person, what they like, your budget, and what can actually arrive on time. Yukti keeps that context together. Users can message it through Linq, import upcoming Google Calendar events, and save details about the people in their lives. It uses Senso and Gemini to research current products, then asks the user to approve the exact merchant, item, amount, and purchase window before opening a Prava checkout. Payment details stay inside Prava and are never stored by Yukti.
SafariGigs started as a mobile-first job marketplace for Tanzania. After building the full job board, I began repositioning it around a bigger hiring problem: screening candidates quickly and consistently. The current direction is an AI interviewing and vetting product for Tanzanian recruitment agencies and employers. Voice or avatar interviews collect structured answers, generate candidate summaries, and help recruiters decide who should move forward. The original marketplace remains live while I validate the new workflow, local voice quality, pricing, and agency demand.
I built InOrbit after noticing that the hard part of networking is not meeting people. It is remembering why they mattered and following up at the right time. On mobile, users can capture a person with a quick note, voice memo, card scan, or badge. On web, they can review contacts, see relationship history, draft follow-ups, and plan next steps. The product brings contact memory, enrichment, reminders, calendar availability, and selected-file sharing into one workflow.
SolveYVR makes it easier for Vancouver residents to report problems without working through a long city form. A user can record a short video, add a voice note, and let the app identify the issue, pull out the useful context, match it to the right Van311 category, and prepare a report. It also places nearby reports on a public map so people can see whether the same problem has already been raised. I built it with Next.js, FastAPI, MapLibre, Convex, vision models, speech-to-text, geocoding, and Vancouver open data.
GhostLine turns a live call into a paranormal containment hotline. The user speaks with an AI operator, points the camera around the room, and completes tasks based on what the agent can see and hear. The agent reacts in real time, challenges weak evidence, and ends the session with a structured case report and containment verdict. I built the experience with Gemini Live on Vertex AI, a React client, and a FastAPI WebSocket backend for the Gemini Live Agent Challenge.
Cascade takes a public GitHub repository and a messy request, then turns them into a visible implementation mission. A user can paste meeting notes, feedback, or a feature idea. Cascade reads the codebase, clarifies the goal, shows PM, architecture, implementation, and QA agents moving through the work, and attempts a real patch for supported JavaScript and TypeScript projects. The final output includes changed files, checks, blockers, and proof of what happened.
Clearfeed is for people who use X to learn and meet others but do not want to live inside the feed. It watches selected lists and timeline sources, ranks the posts most likely to matter, and helps draft replies, quote posts, and original posts. The user stays in control of everything that gets published, and the system learns from what they edit, accept, or reject. I built it as a local-first tool so the writing context and workflow remain under the user's control.
This Python project was an experiment in turning prediction-market data into a repeatable research workflow. It collects changing market information, watches relevant events, and tests how rules could be used to flag opportunities or changes in probability. The focus was on data structure, monitoring, and decision logic rather than presenting it as a finished trading system.
I built this Python workflow to test how much of a research-to-content pipeline could be automated. It gathers facts, organizes them into a clear sequence, and turns the result into repeatable carousel-style posts. The project helped me explore where automation saves time and where human review is still needed to keep the output accurate and worth reading.
Product Ops Studio is a polished demo application I built as a realistic target for repo-aware coding agents. It includes a marketing site, onboarding, pricing, roadmap cards, workflow lanes, and a dashboard-style mission view. The point was to give tools like Cascade a codebase that looked and behaved like a real product instead of testing them against a toy repository.
I designed this platform to make trade review more useful than a spreadsheet of wins and losses. It logs trades, supports strategy backtesting, compares performance across setups, and helps the user look for patterns in risk, timing, discipline, and decision-making. The goal was to turn raw history into feedback that could improve the next decision.
I developed a Django and PostgreSQL inventory system for a textile manufacturer that needed a clearer view of raw materials and finished goods across departments. The system synchronized inventory changes, reduced manual tracking, and made stock discrepancies easier to find. I also wrote the technical and user documentation needed for the team to adopt and maintain it.
This university project used computer vision and deep learning to detect faces and classify visible emotional expressions from images. I worked through the full pipeline, including data preparation, model training, evaluation, and real-time inference, while learning where facial expression labels can be useful and where they can oversimplify human emotion.
I built and trained a convolutional neural network in TensorFlow to classify handwritten digits from the MNIST dataset. The project covered loading and normalizing the data, augmentation, training, evaluation, and documenting the code so the results could be reproduced.
I implemented attention inside a pre-trained sequence-to-sequence translation model to better understand how the model decides which parts of an input matter while generating each token. I documented the method, compared the model's behaviour, reported the results, and outlined the changes I would test next.
A responsive React application that fetches current weather from the OpenWeather API and presents it in a simple interface for mobile and desktop. I built it while learning how to work with public APIs, asynchronous state, error handling, and responsive component design.
A small web app that pulls a new piece of advice from the Advice Slip API whenever the user asks for one. I built it with HTML, CSS, and JavaScript, with responsive layouts and interaction states for both desktop and mobile.
I recreated the core flow of a fashion e-commerce site to practice building a complete shopping experience without a framework. The project includes authentication, product pages, cart state, coupon logic, and a simulated checkout flow using HTML, CSS, JavaScript, and LocalStorage.
A desktop Tic Tac Toe game built with Python and tkinter. It was a small project for practicing event-driven interfaces, game-state logic, and packaging a simple application from start to finish.
I'm Yash Serai, a product engineer, data engineer, and founder. I like taking products from a rough idea through design, backend, deployment, and the point where real users start showing you what is wrong.
My main project is Soulspace, an AI-native fandom social app where fans join worlds around shows, anime, and games. AI characters participate through profiles, feeds, DMs, and group chats, so interacting with them feels connected to a wider community instead of trapped inside one chat window.
I also build smaller products to test specific ideas. Buildmates uses the work a builder chooses to share with Codex to find people worth meeting. Yukti keeps track of the people and occasions in your life, researches a thoughtful gift, and asks for approval before checkout. SafariGigs began as a Tanzanian job marketplace and is now being repositioned around AI interviewing and candidate vetting for local recruitment agencies.
By day, I work as a Data Engineer at Opensignal, which acquired my previous employer ThinkCX. I was the only junior engineer retained through the acquisition. My work has included automated QA across hundreds of business rules, data pipelines for previously unmappable datasets, and reporting systems that remove hours of manual work for clients.
I hold a Bachelor of Science in Computing Science from Simon Fraser University. My background covers data engineering, full-stack development, machine learning, DevOps, and consumer AI. I am most interested in products where the technical system and the user experience have to be designed together.
Languages: Python, JavaScript, TypeScript, SQL, C, C++, R, Java, Kotlin, Matlab, Go
Frontend: Next.js, React, React Native, Tailwind CSS, shadcn/ui, Radix UI, Framer Motion, TanStack Query, React Hook Form, Zod, MapLibre
Backend & Data: Django, Node.js, Express, FastAPI, PostgreSQL, Supabase, Convex, pgvector, Celery, REST APIs, WebSockets, Firestore, Railway, Cloud Run, Cloudflare R2
AI/ML: TensorFlow, PyTorch, Scikit-learn, Gemini, Gemini Live, Vertex AI, OpenRouter, ElevenLabs, Cartesia, Deepgram, Whisper, Numpy, Pandas, Matplotlib
DevOps & Tools: Git, Docker, Linux, Bash, Playwright, Firebase, Clerk, Resend, JUnit, SQLite, Microsoft Azure
Certifications: Microsoft Certified: Azure Developer Associate | Microsoft Certified: Azure AI Engineer Associate
Joined ThinkCX as a Data Engineer and was selected as the only junior engineer retained following Opensignal's acquisition of the company. Deployed an automated QA system in Python and SQL that validates daily data against 700+ business logic flags. Improved data accuracy by identifying and resolving a critical data integrity flaw in a key client dataset. Engineered a specialized data pipeline using caller ID to map and ingest data from previously unmappable network ports, expanding usable datasets. Developed automated reporting infrastructure that generates and distributes data visualizations for clients, reducing manual reporting work.
Delivered personalized product recommendations and product demonstrations in a fast-paced retail environment. Maintained strong customer satisfaction by adapting explanations to different user needs and collaborating across store teams.
Monitored and optimized server performance, reducing system downtime and improving station operations. Developed technical documentation for configurations and troubleshooting procedures, assisted with network infrastructure setup, and implemented automation scripts for routine maintenance tasks.
Identified and resolved software bugs, supported development of a real-time inventory tracking system, and authored documentation for inventory management workflows. Contributed to a system that improved stock visibility and reduced discrepancies across departments.
Interested in consumer AI, social products, developer tools, data engineering, or working together? Send me a message.