hi i'm chirag.
i build ai products that feel simple in the browser and serious under the hood.
been here since 23.000000000 years.
software engineer at HSBC in Pune.
i like LLM systems, retrieval infrastructure, security products, and tools that make developers faster.
currently building Phishly,
contributing to Agent Orchestrator,
and trying to make my side projects less like demos and more like products.
i probably talk to codex more than most humans. powered by coffee, math, python, and the occasional unreasonable weekend build.
now.
- building Phishly into a fast, useful security product for small teams.
- contributing to agent orchestration infra where reliability actually matters.
- learning how to make AI systems feel less magical and more inspectable.
- keeping a small collection of things that sharpen my taste.
what i care about.
i am drawn to systems where the interface looks obvious, but the backend is doing real work:
retrieval, orchestration, inference, realtime events, safety checks, and weird edge cases that only show up after shipping.
the goal is not to look busy. the goal is to build things that make a technical person pause and think,
"okay, this person actually builds."
stuff i've built.
Phishly
the B2B SaaS product i'm currently building: an AI-native phishing simulation platform for startups and small teams. companies can safely simulate real attacks, identify vulnerable employees, and train them before a real attacker does.
Composio Agent Orchestrator
open-source work on a platform that coordinates fleets of AI coding agents across isolated worktree, branch, and PR flows. i shipped state/session reliability fixes, CLI JSON/watch output, and dashboard/doctor tooling.
13 PRs merged
VecSearch
a vector database built from scratch for semantic search and retrieval. embedding storage, similarity indexing, cosine search, and the kind of internals you understand better after rebuilding them.
feb 14, 2026
PaperLens
turns dense research papers into visual explanations and structured insights. built for the moment when a paper is important, but your brain needs a map before it needs more text.
feb 22, 2026
PocketLlama
an on-device LLM implementation using TinyLlama. offline conversational AI, mobile inference constraints, and a useful reminder that models are only one piece of the product.
jan 21, 2026
SkyPing
realtime flight tracking around Pune Airport with event-based notifications. small obsession, real system: live data, filtering, and signal instead of noise.
nov 21, 2025
API-Talks
a conversational layer over API documentation so developers can ask questions naturally and get context-aware answers without spelunking through static docs.
nov 2, 2025