# Sanjay Krishna Anbalagan

> Senior Engineer @ AWS · PhD in Computer Science. Builds open-source developer abstractions that make enterprise generative-AI applications explainable by construction.

The canonical machine-readable view of this site is available at:
https://sanjay1909.github.io/?view=source

It can be served as JSON, YAML, or MDX via the `format` query parameter.

## Identity

- Name: Sanjay Krishna Anbalagan
- Role: Senior Engineer at Amazon Web Services (AWS)
- Education: PhD, Computer Science, University of Massachusetts Lowell
- Username: sanjay1909 (GitHub, Medium, LinkedIn)
- Homepage: https://sanjay1909.github.io/

## Thesis

Explainability is a property of the substrate, not a feature layered on top. If a backend can produce a causal trace of every decision it made, the same artifact is debuggable by engineers, reasonable-over by models, and auditable by regulators. This portfolio is itself an instance of that thesis — the page is narrative, index, graph, and machine-readable source at the same time.

## Journey (the "About" section)

I have been shipping LLM-backed applications since the early days of GPT-3, growing alongside the models through every era. Four eras, four lessons:

1. **2020 — Highlight era (baby bot).** Wrap every important fact in `<highlight>` tags or the model forgot it before the end of its own sentence.
2. **2022 — Chain-of-Thought era (toddler).** "Think step by step." One rung at a time.
3. **2023 — Tree-of-Thought era (high-schooler).** Generate three candidate plans, score each, prune the rest.
4. **2024 → now — Orchestration era (junior employee).** Tool graphs, ReAct, RAG, routing, evaluation per iteration. Emit a causal trace so the audit passes.

Every prompt-engineering hack we invented to teach the model became baked-in instinct in the next generation of weights. The prompts died. The shape they were compensating for did not — a graph of decisions, each with a cause. **FootPrint** is that graph, honest. **agentfootprint** is what happens when you let the graph host an LLM as one of its operators. Both are the condensed form of what a decade of raising these applications taught me. Companion essay: https://medium.com/data-science-collective/llms-from-baby-bots-to-expert-employees-e99aa5553ed6

## Projects

- [FootPrint](https://github.com/sanjay1909/footprintjs) — 2024 — TypeScript framework that turns backend business logic into a directed graph and produces causal execution traces. `npm install footprintjs`.
- [agentfootprint](https://github.com/sanjay1909/agentfootprint) — 2025 — TypeScript framework for explainable AI agents, built on FootPrint. Supports LLM Call / ReAct / RAG / Sequential / Parallel / Routing patterns with per-iteration evaluation. `npm install agentfootprint`.

## Papers

- "Bridging UI Design and Chatbot Interactions" — HCI International 2025, Springer Proceedings.
- "Visible Reasoning" — HCI International 2026, Springer Proceedings (accepted). A framework for deterministic LLM agent transparency.

## Writing

Newsletter: *Enterprise Gen AI Application* on LinkedIn, 320+ subscribers.

- № 01 · From Supply-Driven to Demand-Driven — the chatbot should drive the UX, not assist it.
- № 02 · Make Search the First Tool — STAY/SWITCH plus a Focus Token for multi-turn agent sessions.
- № 03 · The Flowchart Pattern — why backend code should be designed as a graph the model can read.

## Contact

- LinkedIn: https://www.linkedin.com/in/sanjay-krishna-anbalagan/
- GitHub: https://github.com/sanjay1909
- Medium: https://medium.com/@sanjay1909