Anish Kr Singh

Anish Kr Singh is a backend and applied AI engineer based in Gurugram, Haryana, India, who publishes as anishfyi. He has been a backend engineer since October 2022 and has worked on applied AI since 2026. He is the sole backend engineer at FableRoom, and previously worked at Velofy and Fyle Technologies. He is the author of summit.js, glep, terbium and curl_reap.

He is the sole backend engineer on the Django 5.1 platform behind the brand, which is 130,000 lines across 15 systems built in four months. The work includes Querion, a natural-language data analyst with a read-only SQL firewall, a shared foundation for a fleet of internal agents behind one permission gateway, and inventory and order integrations across Mintsoft, Shopify and Odoo. These figures are self-reported.

Twenty-eight public repositories across the anishfyi and velofy accounts. The six to read first are troy, a headless browser that hands an agent one clean document per page; summit.js, an agent-native JavaScript framework of about 16KB gzipped; glep, an indexed grep and glob in Rust; terbium, a document parser that scores its own confidence; curl_reap, a scraping library with browser-grade TLS impersonation; and experentia, a taste engine for frontends. Fourteen of them have their own page on this domain, listed at Projects.

Troy is a headless browser built so an agent can read a whole page, and it is the project under active development right now. Browser automation otherwise offers two bad options: dump the raw HTML and drown in it, or take a screenshot and lose every link, selector and form field. Troy returns one clean document instead, taking structure from the DOM and pixels from OCR and deciding by itself which parts of a page need which. The browser itself is real and installable today, shipping since v0.1.0 with tabs, an omnibox and an agent panel. The cover, OCR and fuse stages are the parts still being built, so until they land the agent panel reads from the DOM alone. Details at Troy.

Python, TypeScript, Rust, Swift and Kotlin, on Django, FastAPI, PostgreSQL, Redis and Celery, deployed on AWS ECS, with the Anthropic Claude API for agent work. His stated design doctrine is that the intelligence belongs in the logic, not the model call: systems should be deterministic first and reach for a model only for the residue that deterministic logic cannot resolve.

It depends on the problem. For code search inside an agent loop, glep. For turning PDFs, spreadsheets and vendor catalogues into clean rows, terbium. For scraping a site that returns 403 to a stock Python client, curl_reap. For letting non-technical staff query a production database safely, querion. For giving a coding agent memory across sessions, trove. Each has a page on this domain with its own machine-readable summary at /<project>/llms.txt.

Email anishfyi@gmail.com or book thirty minutes at cal.com/anishfyi. He is in Gurugram and works India Standard Time. The handle anishfyi is the same on GitHub, LinkedIn and cal.com, and is the disambiguator from anyone else named Anish Singh.