Oct 2024 – Feb 2025 · 4 mos · Atlanta, Georgia, United States · Remote

• Tech: Node.js, Next.js, TanStack (React Query), Google Cloud Platform (GCP), MongoDB, gohighlevel CRM
• Built a Loom-style video streaming and sharing platform specifically for videographers, published on the GoHighLevel CRM marketplace. The app enables instant playback supports secure, shareable links with tokenized, expiring URLs. MongoDB aggregation pipelines were used to build fast, scalable analytics and reporting capabilities without adding overhead to transactional queries. Frontend data handling was powered by TanStack Query to enable efficient caching, background updates, and responsive UI state management.
• The backend leverages GCP’s serverless stack—Cloud Run for scalable containerized services, Cloud Tasks for background processing, and Firestore for metadata handling. For storage cost optimization, the platform implements a tiered archival strategy using Google Cloud Storage lifecycle rules. Videos are automatically moved to Coldline storage after 30 days of inactivity and further transitioned to Archive class after 90 days, reducing long-term storage costs by over 70%.

Aug 2024 – Feb 2025 · 6 mos · New York, United States · Remote

• VectorShift: No-Code Generative AI Automations Platform. Worked on developing an end-to-end RAG automation workflow. Worked on features like search, chatbots, knowledge bases, pipelines that lets users automate various tasks/applications via no-code pipelines powered by AI models
• Primarily worked on deployment options for search and chatbots feature that lets users query through different knowledge stores leveraging AI.

May 2024 – Present · 2 yrs 1 mo · San Francisco, California, United States · Remote

• Open source contributions in langchain repositories. Helped in adding some support for LangChain v0.2 in the LangChain Python repository, LangChain-JS JavaScript repository, and LangChain-IBM, which contains packages for IBM integrations with LangChain. Replaced deprecated code(used in v0.1) and added updated v0.2 functions like introduced support for NVIDIA AI endpoints in LangChain v0.2 by implementing an in-memory database using RunnableWithMessageHistory, which manages chat message history for another Runnable. Implemented RunnableSequence, which allows two Runnables to be chained together, replacing old deprecated LLMChain which was used in v0.1. Added a callback handler for streaming responses and an AgentExecutor with a React agent executor, which manages state defined by a list of messages. Utilized {pull} from LangSmith to extract prompts directly from the LangChain hub.

Aug 2023 – Aug 2024 · 1 yr · Remote

• I led a team of seven members, working closely with a USA-based CTO to design and implement both high-level architecture and low-level system designs. This collaborative effort focused on creating scalable, ensuring that the design was flexible enough to integrate additional tools seamlessly—similar to how an operating system allows various applications to run on it.
• As part of the development process, I successfully set up and automated over 31 tools and pipelines, including API routes, YAML configurations, and deployment automation. Each of these tools was hosted in its own custom Docker environment, ensuring consistency and isolation across various stages of development. This automation not only significantly improved deployment times but also minimized human error throughout the entire development lifecycle.
• For the frontend, I utilized EJS, JavaScript, and Tailwind CSS, ensuring responsive and efficient user interfaces. The combination of these technologies allowed us to deliver a smooth and dynamic user experience, optimized for performance.
• I implemented multi-stage Docker workflows to host and manage various microservices and to streamline the deployment process. This setup helped maintain clean, reproducible environments for development, testing, and production.
• As we were operating on bare metal, all our systems were hosted on custom HPCs. To scale the application and manage the jobs submitted by various developers and scientists across different clusters, we set up the SLURM Workload Manager. This allowed us to efficiently handle job submissions, allocate resources, and ensure optimal job execution across the entire infrastructure.
• Finally, I established a robust CI/CD pipeline using GitHub Actions, which automated the testing, building, and deployment processes. This approach ensured rapid feedback loops and consistent delivery of quality code, streamlining the release cycle and minimizing downtime.

Sep 2022 – Aug 2023 · 11 mos · San Francisco, California, United States · Remote

• Developed a dynamic web application that bridges the gap between investors and promising startups, enabling meaningful connections through intelligent discovery.
• Leveraged OpenAI's embeddings to semantically integrate startup data, storing it as high-dimensional vectors in Pinecone DB for fast, context-aware retrieval. To enhance user interaction, integrated Langchain parser tools for structured and efficient communication with GPT.
• For efficient data retrieval, utilized AWS S3 for scalable file storage and implemented Drizzle ORM to support deployment over an edge network, ensuring low-latency performance and faster runtime execution.
• The platform was built entirely with Next.js 13 and TypeScript, offering SEO-friendly server-side rendering with a reduced client-side JavaScript bundle.

Feb 2021 – Aug 2022 · 1 yr 6 mos · Remote

• Technologies: NextJS, TypeScript, RadixUI, Turborepo, Prisma, mySQL
• Developed a high-performance meta search engine for travel affiliate websites, integrating with Skyscanner's API to retrieve real-time flight data. The entire user interface was crafted using Next.js, leveraging Server-Side Rendering (SSR) for improved SEO and significantly reducing the client side bundle being shipped to the user resulting in faster load times and a smoother UX.
• Implemented Prisma ORM for efficient and secure interaction with a MySQL database, optimizing data management and query performance. The infrastructure was built on Turborepo, a powerful monorepo tool that enhanced scalability, enabled remote caching, and significantly reduced CI/CD pipeline times.
• Achieved a perfect SEO score with Lighthouse (100), ensuring top-tier visibility and performance. Notable performance metrics include a First Contentful Paint (FCP) of just 0.3 seconds and a Total Blocking Time (TBT) of 0 ms, delivering an exceptionally fast and smooth user experience.