Journal of record Prabakaran Chandran · est. 2024 · Tamil Nadu → New York
PraCha

Prabakaran Chandran  ·  பிரபாகரன் சந்திரன்

Edition Morning Edition · press the seal
Vol. III · No. 251   New York Morning Edition

The frame

  • Everything is a system.Control theory gives you something: anything that can be seen as a system can be modeled, understood, acted upon.
  • Building is becoming.Not building to show something — building as the way the thinking gets real and the person gets formed.
  • I stopped wanting to settle.Industry, under time pressure, lets you settle for “good enough.” That is the gap I keep walking back into.
  • From sensors to tensors.Control engineering, then complexity, then data, then machine learning, then the research underneath all of it.
  • The work is what matters.I am not chasing the usual signals. I know what I have built and what I am building toward. Manhattan is temporary.
By Prabakaran Chandran Tamil Nadu → New York — From sensors to tensors
Prabakaran Chandran
Prabakaran Chandran · பிரபாகரன் சந்திரன்

I chose control engineering somewhat by accident. My plus-two marks weren’t what I needed for the top courses: mechanical, computer, triple E. Control engineering was offered in just seven colleges across Tamil Nadu. Less competition, I thought. Maybe I’d get a gold medal. That was the whole plan. It was naive. I didn’t get the medal either. Health issues got in the way. But what I did get, without looking for it, was a lens. Control theory gives you something: everything is a system. Anything that can be seen as a system can be modeled, understood, acted upon. I’ve never stopped seeing the world that way.

Then came Mu Sigma. Over three years there, the frame kept widening: systems, behavior, intelligence. After that: Captain Fresh, Informatica, a stealth startup. Six-plus years of production ML systems: satellite imagery, aquaculture, document understanding, enterprise AI, agentic pipelines. I learned how to operate. I learned where the real friction is. I also noticed what I kept running into: the gap between knowing how to run a model and knowing why the model should work the way it does. Industry, under time pressure, lets you settle for “good enough.” I stopped wanting to settle.

“I didn’t get the medal either.”On choosing control engineering · continued in About

I’m not chasing the usual signals. I know what I’ve built, I know what I’m building toward. Building is the mechanism of becoming, not building to show something, but building as the way the thinking gets real and the person gets formed. Each new idea worked on is a small act of self-construction. The work is what matters. Manhattan is temporary.

Continued in D · About →

Building right now

Three pillars.

日月

  1. Independent lab — Downstream of Intelligence

    PrachaLabs

    An independent, one-person lab, downstream of intelligence. Intelligence is becoming abundant, and the opportunity is not to automate what we already do but to discover what we can now attempt. Upstream, how frontier intelligence is made; downstream, how real value gets created with it. Research, engineering, product, and strategy as one motion.

    Enter the lab ↗

  2. Frontier AI — Research

    Research Engineering

    Frontier AI research engineering, run as sprints: each begins with a precise question and ends with working code, a finding, or a documented failure. The experiments, deep dives, and independent research that move from reading the frontier to building at it.

    See the frontier →

  3. Decision Engineering — −1 to 0

    Evolving Decision Systems

    The right answer today is quietly the wrong one tomorrow, yet AI is frozen the moment it ships, and we cover the gap with armies of people rebuilding systems by hand. We’ve gotten good at building AI that’s smart once. Evolving decision systems from the intersection of AI, complex dynamical systems, and adaptive control — for clinical trials, health, life sciences, finance, and operations.

    Explore the thesis →

A · WorkBuilds · focus sprints · PraCha, 8 Sept 2026

AWork

Each sprint begins with a precise question and ends with working code, a finding, or a documented failure.

2025 – ongoing — Agentic wiki · Frontier AI

FAIRE — Frontiers in AI Research and Engineering

An agentic frontier AI wiki that turns scattered learning into directed building: 10 canonical tracks, concept pages, primary-source references, and Minimum Valuable Builds that push each topic toward implementation. Self-designed and self-directed. Every sprint produces a public artifact.

2025–2026 — Reinforcement Learning · Multi-Agent

Zero-Shot Coordination in Multi-Agent RL

Agents trained separately fail to coordinate when paired with a stranger — each one built different habits. The question is whether coordination can be learned as a general strategy rather than a shared convention, so that meeting a stranger stops being a special case.

2025–2026 — Continual Learning · Causal Inference

No-Exemplar Continual Learning via Causal Invariance

Neural networks forget old tasks when learning new ones. Instead of storing past examples, this asks why forgetting happens — and finds the answer in causal structure. Outperforms replay-based baselines with zero stored images.

2026 — Physics-Informed ML · Dynamical Systems

Energy-Based Port-Hamiltonian Neural Networks

Neural networks predict physical motion but do not know that energy is conserved, so over a long horizon they drift. Here conservation is baked into the structure rather than left as a soft constraint the model is asked to learn.

2025–2026 — Causal Inference · Time Series

Causal State-Space Models for Time Series

Forecasting models predict what will happen — but decisions change the pattern. This integrates causal structure into sequence models so they can reason about interventions, not just correlations.

2026 — Causal Representation Learning · Recommendation

HMV-CRL: Separating Platform Influence from Genuine Preference

Recommendation engagement conflates what users want with what the algorithm pushed. This model learns both representations separately — and finds that platforms amplify engagement while suppressing the genuine preference signal.

2025 — LLM Agents · Social Simulation

Generative Social Network Simulation

Social simulations use hand-coded rules. This uses language models as the agents themselves — each with distinct beliefs and tendencies — to study how opinions spread and polarize across a network.

2025 — Vision-Language Models · Alignment

Personality-Aligned Vision-Language Model

People describe the same image differently depending on how they think. This aligns a vision-language model to specific personality styles, then tests whether that consistency makes it a better coaching assistant.

Ongoing — Research Engineering · AI Systems

Engineering Practice Log

A running place for the engineering I want to compound for the future of AI, ML, and data systems: evaluation, agents, tooling, data systems, inference, reliability, and end-to-end system design.

All builds and focus sprints · Section A →

B · WritingDaybook · Musings · Library

BWriting

What I read, how I am changing, people and institutions I learn from.

Featured · mindset · 1 July 2026

A State of Me, and What's Next

How the work kept scaling — from a single dataset to shaping how systems think and decide — what changed in me along the way, and what comes next.

Read →

Library

Musings

All 55 essays · Section B →

C · ThesisDecision Engineering · FAIRE · Curriculum

CThesis

Evolving Decision Systems. −1 to 0.

Decision engineering — systems that keep deciding well after they ship.

AI, control science, and complexity science are not three separate tracks. They converge. Control gives you the language of state, dynamics, and feedback. Complexity science shows what happens when those go nonlinear: emergence, adaptation, surprise. AI is the machinery that makes both tractable at scale.

The thesis and the work · Section C →

FAIRE — the research program

Self-designed and self-directed, sprint-based, at the frontier of AI research and engineering. Four objectives as a loop, not a sequence: Discovery, Evidence, Inference, Optimization. Every sprint produces a public artifact: code, a writeup, a failure post-mortem. Nothing stays private. The accumulation is the portfolio.

FAIRE → · The frontier →

Curriculum — eight tracks

  • 01AI
  • 02Generative Modeling
  • 03Representation Learning
  • 04Neural Networks & Deep Learning
  • 05Statistical & Probabilistic ML
  • 06Causal and Statistical Inference & Modeling
  • 07Algorithms and Systems for AI
  • 08Complex Dynamical Systems, Control and Discovery

Curriculum →

D · AboutSystems → Data Science → ML Engineering → AI Research → Decision Engineering

DThe Arc

M.S. Data Science, Columbia, December 2026 · B.E. Control Engineering, Anna University.

I
Systems & Complexity

Control engineering taught me to see everything as a system — state, dynamics, feedback. Mu Sigma widened that to complexity: emergence, nonlinearity, how simple rules produce intricate behavior.

II
Data Science

Six years turning data into decisions — not reports that sit in decks, but models that organizations actually act on across industries.

III
ML Engineering

Satellite imagery, aquaculture, document understanding, enterprise AI. Systems that shipped, scaled, and changed how real operations ran.

IV
AI Research

At Columbia: filling the theoretical gaps industry doesn’t have patience for. Causal reasoning, continual learning, reinforcement learning, probabilistic modeling.

V
Decision Engineering

The destination. Where everything converges — AI, causal thinking, complexity science — into systems that change how decisions get made at scale.

Full story · Me → · Résumé →