Susankar Karmakar has launched Susan AI, a personal assistant project that rejects the standard chatbot paradigm in favor of a unified digital workspace. Deployed on Vercel and powered by Google Gemini Flash-Lite, the tool is designed to handle complex, multi-step tasks like research synthesis, code debugging, and document analysis. Karmakar describes the project as an ongoing experiment in combining generative AI with productivity tools, aiming to move beyond simple question-and-answer interactions toward true workflow automation.

Architecture and Core Philosophy

The technical foundation of Susan AI relies on a modern web architecture that integrates an AI application layer with user data and external tools. Karmakar explicitly states that the goal is to transform the typical 'Question β†’ Answer' loop into a more robust pipeline: 'Goal β†’ Context β†’ AI reasoning β†’ Tools β†’ Output β†’ Workflow.' This shift requires the system to maintain context across different tasks, allowing users to start with research, move to writing, and finish with coding without losing the thread of their work.

From Chat to Actionable Workflows

Current capabilities include AI chat, learning assistance, writing support, coding help, and document processing. The project distinguishes itself by attempting to bridge these silos. For instance, in the coding module, the workflow moves from problem explanation to code generation, review, error finding, and implementation improvement. Karmakar is also experimenting with agent-style workflows where the AI plans, uses tools, processes information, and takes actions autonomously, rather than waiting for sequential user prompts.

Lessons in AI Engineering

Building the assistant has highlighted that connecting an LLM API is merely the starting point. Karmakar notes that a useful AI product requires rigorous UX design, precise context management, and robust error handling for API failures and rate limits. He emphasizes that 'AI should assist the writerβ€”not replace the writer,' advocating for human control over generated content. The project also stresses the importance of trust, noting that confident AI answers are not necessarily correct, especially in research or decision-support contexts.

Key Takeaways

  • Susan AI is built on Google Gemini Flash-Lite and deployed via Vercel, emphasizing accessibility and modern web standards.
  • The project aims to replace isolated chat interactions with integrated workflows for research, writing, and coding.
  • Karmakar identifies context management and graceful error handling as critical engineering challenges beyond basic model integration.
  • Future development focuses on better AI memory, tool calling, and voice interaction to create a true personal digital workspace.

The Bottom Line

Susan AI is a compelling reminder that the next frontier in dev tools isn't just smarter models, but better orchestration. Karmakar’s focus on workflow integration over raw chat capability aligns with where personal AI is actually heading: from novelty to necessity.