What I have built
I build AI agents, automate finance and reporting work, and build the platforms they run on. I lead developers and teach teams to use AI. I have worked as an independent contractor since 2019. I taught myself to program through CS50, Harvard's free online course, and I have an MBA from the Indian School of Business.
Most of my work is for finance teams: agents that chase unpaid invoices, review expenses and answer revenue questions in chat, reporting packs and board decks that build themselves, and the setup and training a team needs to use AI safely.
I came to this from the business side: e-commerce at Fabindia, then corporate finance at Virtusa, where I moved into automation in 2019.
Outside work I plant trees on a farm and build automatic watering for them. I also make small apps for myself and websites for friends, and I surf.
Professional work
Described in general terms. Ask me for details.
Finance operations automation
- A collections platform with dashboards and AI agents that help a finance team follow up on unpaid invoices (built with a team)
- Agents that sort incoming replies and draft customer emails for a person to review
- Agents that turn meeting and call transcripts into follow-up records
- Browser and document automation for timesheets and invoicing
- An expense review platform that reads claims and receipts with document AI, checks them against policy with an LLM workflow and leaves approval to a person
Technology: Next.js, React, TypeScript, Python, FastAPI, LangGraph, Azure OpenAI, Azure Document Intelligence, Playwright, Microsoft Graph.
Chat agents for business data
- Chat agents inside Teams and other workplace chat tools that answer revenue and KPI questions in plain language, with the arithmetic done by code rather than by the model
- An analysis web app where the model builds tables on demand next to the conversation
- An assistant that answers only from approved material and declines everything else
- MCP servers that expose analysis tools to any compatible agent
Technology: Python, LangGraph, LangChain, Azure OpenAI, Gemini, Vertex AI, MCP, FastAPI, Streamlit, Bot Framework, Cloud Run.
Reporting automation
- Generators that build management and board decks straight from workbooks and check every slide against the source data
- Pipelines that produce recurring reporting packs in multi-sheet Excel
- Weekly cost and margin packs assembled from APIs instead of by hand
- Power BI reports, tabular models and DAX measures
Technology: Python, pandas, python-pptx, OpenXML, openpyxl, Power BI, DAX, VBA.
Workforce and cost analytics
- A staffing KPI dashboard with data uploads controlled by role
- Pipelines for headcount, compensation and people-cost forecasting
- A simulator that shows how policy changes affect cost and margin
Technology: Next.js, TypeScript, Python, pandas, Flask, Google Cloud Storage, Power BI.
AI platform and training
- An internal catalogue where teams find and install agents, skills and MCP servers
- A registry of MCP servers, plus server templates with sign-in and container deployment
- Reusable skills for coding agents, such as masking data before a model sees it, pulling ERP and HR data, building slide decks and drafting email
- A platform where business users run their own scripts behind a code-safety check, and an agent explains and repairs failures
- Cloud cost guardrails and budget alerts for AI workloads
- A security review of internet-facing apps, run with a coding agent
- A self-hosted deployment of an open-source agent with a chat gateway
- Handbooks and setup guides for teams adopting AI models and coding agents
Technology: MCP, Python, FastAPI, Azure Functions, Azure Container Apps, Azure AI Foundry, Vertex AI, Docker, Bicep, PowerShell.
Programme and project management
- Running a finance AI programme, with tooling I built to track its use cases, commitments and follow-ups
- Planning and tracking tools for a team of developers
- Sorting a backlog of proposed AI agents into true agents and ordinary automation
- Specifications, process maps and delivery boards for planned agents
Technology: Python, Flask, Azure DevOps, BPMN, Power Automate.
Data engineering
- Pipelines that turn large Excel workbooks into validated, queryable datasets with historical snapshots
- Data extraction from a cloud data warehouse and from Spark
- No-code tools that let finance users define and run their own data consolidation
Technology: Python, pandas, Parquet, SQL, PySpark, Databricks, Apps Script, React.
Small tools
- Meeting transcription and summaries that run on my own machine, keeping recordings off external services
- A process chart viewer and a calendar sync utility
- A rate prediction model and a knowledge graph of operating procedures (prototypes)
Technology: Python, Whisper, Svelte, scikit-learn, Neo4j.
Earlier career
- Fabindia, a craft and lifestyle retail brand (2009 to 2013): managed the technology side of its e-commerce, helped set up the project that built the mobile site, sold through online marketplaces and automated spreadsheet work; also executive assistant to the managing director
- Virtusa, a global IT services company (2014 to 2023): in corporate finance until 2019, where I brought an acquired company's financial data into the parent's systems, shortened purchase-to-pay and order-to-cash cycles and worked in financial planning and analysis; from 2019, as an independent contractor, automated its management reporting
- Rangsutra Crafts, a craft social enterprise (2019 to 2021): ran its online store, payments, shipping and operating procedures
- Yuvabe (2022 to 2024): worked as tech lead and full-stack engineer for an early-stage startup
Personal projects
Projects outside work, for myself, friends and small local organisations. Several are prototypes.
IoT and home automation
- An automated watering system for my trees: an ESP32 and a solenoid valve that opens on a message over the network and shuts itself off after a set time
- Experiments in soil-moisture sensing on a microcontroller: reading and calibrating the sensors and bench-testing a servo
- A phone lockbox guarded by an AI assistant that asks for a reason before unlocking and logs every unlock (prototype)
Technology: ESP32, Arduino, MicroPython, MQTT, MCP, Python, SQLite.
Farm and community projects
- An interactive 3D map of a farm, built from a land survey, where you can walk around and tap a tree to see what it is
- An organic farm's website, rebuilt several times over the years
- A registration and produce-sales app for farmers (prototype)
- An agent that answers questions about government schemes in Indian languages, drawing on a knowledge base and web search (prototype)
- An introductory AI session for Tamil-speaking facilitators, with an English-Tamil voice translator app
- A calculator that turns a carbon figure into everyday equivalents (prototype)
Technology: three.js, Next.js, Astro, React, Django, FastAPI, LangGraph, Flutter.
Personal apps
- A buy-in tracker for home poker games that works out who pays whom at the end of the night
- A daily habit tracker with streaks
- A pipeline that moves heart-rate variability and sleep data from my phone into a database, where an AI coach reads it to write my weekly training plans
- A goal tracker that checks tasks against the time actually logged
- A command-line tool that helps a group of friends agree on a movie
Technology: Next.js, React, TypeScript, Firebase, Python, SQLite, Flask, Rust, iOS Shortcuts.
Websites for friends and small organisations
- Websites for a surf school, a theatre artist, a yoga teacher, Breakneck (an adventure-travel group) and an impact-measurement consultancy
- The RSVP site for my own wedding
- An impact page on the Rangsutra Crafts online store
- LeMiRide and LeMiWander, a booking app and a travel site for a bike-rental and travel startup (prototypes)
- A stock-management app and store scripts for a small retail business
Technology: Astro, Svelte, Gatsby, Next.js, Vue, Contentful, Shopify, Django, Flutter.