I'm Abhishek Banerjee, a data and AI architect in Toronto. For nine years I've built data systems and, more recently, AI agents — including one that serves 150+ people at the company where I work. I take on a small number of outside projects each quarter: fixed scope, fixed price, and everything runs in your own accounts.
Each one is a defined piece of work with a price we agree on before I start. No hourly meter, no open-ended discovery.
If your team answers the same questions every day from a wiki, a helpdesk, or old support tickets, I'll build an assistant that reads those sources and drafts the answers. It says where each answer came from, and it hands off to a person when it isn't sure.
Most small teams on Google Cloud or AWS pay for things they don't use: machines left running, database tiers far bigger than the workload, storage nobody opens. I go through the bill line by line against actual usage and give you a list of fixes, ordered by how much each one saves. I do the easy ones while I'm in there.
If your numbers live in spreadsheets and nobody quite believes the dashboard, I'll rebuild the shaky parts: automated pipelines into a proper warehouse, tables modeled so each metric means one thing, and checks that catch bad data before anyone sees it.
For teams that need experienced judgment but can't hire for it. I review architectures, help you choose between vendors, and sit in on interviews for data hires. Mostly async, with a call when something needs talking through.
Client work stays private, so here are two things I can talk about. Happy to go deeper on a call.
At my day job I designed the retrieval architecture behind a production AI agent: it answers questions from internal documents, shows its sources, and routes the hard ones to people. It's been running in production with monitoring and regular evals — the unglamorous parts that decide whether these systems survive contact with real users.
I publish reusable skills for building AI agents — including one for deploying retrieval-based agents on Google Cloud — so other teams don't have to start from zero. They're free and MIT-licensed.
I'm Abhishek Banerjee. By day I'm Sr. Manager of Data Platforms and Architecture at Randstad Enterprise in Toronto. I've spent nine years in data engineering across consulting and industry, and the last few years building AI agents that run in production — not demos.
I work mostly in Python on Google Cloud and AWS, with a lot of hours in Vertex AI, LangChain and LangGraph, and MCP-based agent setups. I'm certified as a Google Cloud Professional Data Engineer and a Google AI Professional, among others.
I write about AI engineering on Medium and publish open-source tooling on GitHub. I take on a few consulting projects at a time and do the work myself — you won't be handed to a junior team.
Thirty minutes, free. You describe the problem; I tell you honestly whether I'm the right person. If I'm not, I'll say so and point you elsewhere.
A short written document: what I'll build, how long it takes, what it costs. One fixed price. You approve it before anything starts.
You see a working demo every week, against your real data. If something's off, we catch it early, not at the end.
It goes live in your accounts, with documentation and a walkthrough. I stay available for a support window after handover.
A few lines about the problem is enough. If it looks like a fit, we'll set up a call. If it doesn't, I'll tell you that too.