# Harsh Virani

> AI engineer who builds agents that actually work, and ships them as products.

Open to working-student and full-time AI roles

## Profile

- [About](https://www.harshvirani.dev/about): Background, languages and current focus
- [Email](mailto:harshvirani.91@gmail.com)
- [LinkedIn](https://www.linkedin.com/in/harshvirani914/)
- [GitHub](https://github.com/HarshVirani914)
- [CV](https://cdn.sanity.io/files/h014bgtz/production/0a6b3e20682ad99fe0e18b1c99787ee76f5ae011.pdf): PDF

## Work

- [Flag & Fill](https://www.harshvirani.dev/work/flag-and-fill): Fills every customs field, flags the ones it is unsure of.
- [PocketPulse](https://www.harshvirani.dev/work/pocketpulse): A calm expense tracker, live at pocketpulse.me.
- [Support agent](https://www.harshvirani.dev/work/support-agent): Answers only when it can cite a source. Otherwise, silence.
- [SSRI platform](https://www.harshvirani.dev/work/ssri): An assessment product with real users and real payments.

## Research

- [Agents in the Grid](https://www.harshvirani.dev/research/agents-in-the-grid): Deep-research LLM agents for office work: can an agent do real spreadsheet work reliably, and does that depend more on the model or on the harness around it? A planned study on SpreadsheetBench.
- [The evolving ecosystem of large language models](https://www.harshvirani.dev/research/evolving-ecosystem-of-llms): Decoder, encoder and encoder-decoder models compared across understanding and generation, with prompting, hallucination and code LLMs.
- [AI-augmented low-code and no-code platforms](https://www.harshvirani.dev/research/ai-augmented-low-code): How LLM agents bridge engineers and citizen developers: task-technology fit and real workflow integration.

## Optional

- [Full profile](https://www.harshvirani.dev/llms-full.txt): All of the above in one file
- [Deutsch](https://www.harshvirani.dev/de): The site in German