LinkedIn

CTO at Tulifo · Chennai, India

Arun Kumar Jayakodi

Technology leader in AI and cloud. 18+ years building products and the engineering teams behind them — still hands-on in code.

Open to AI advisory, speaking and select collaborations

Connect on LinkedIn How I build with AI ↓
Portrait of Arun Kumar Jayakodi
01

At a glance.

I've spent 18+ years building cloud-native products and the engineering teams behind them — and I still write the code.

Today I'm CTO of Tulifo, where I architected and built Xiya, our interview model, six products on one model family and the open-weights Chittu models. Before that I led a 45-person engineering organisation at Watermark Insights with people in the US and China, and designed cloud platforms at Francium, Accenture and Wipro across India, Dubai, Singapore and Spain.

Years building
18+
Largest team led
45
Countries worked across
6
Fewer production bugs
70%
  1. CTOTulifo · India
  2. Engineering ManagerWatermark Insights · US, remote
  3. Technical ArchitectWatermark Insights · US, remote
  4. Solutions ArchitectFrancium · India, Dubai, Singapore, Spain
  5. Technical ArchitectAccenture · India
  6. Senior Technical LeadWipro · India
  7. AWS Certified Solutions Architect – Professional · IIM Bangalore · BITS Pilani
02

From customer problem to model in production.

How I build with AI — four habits that took Tulifo from a blank repo to six products and two open-weights models.

01 · Forward deployed engineering

Embedded with the customer, accountable for production.

I do my best work inside the customer's meetings, codebase and constraints — turning a fuzzy problem into software that runs.

  • Took Tulifo from a blank repo to a live production platform in three months.
  • Built fintech and mobility platforms for MoneySmart and Spin with teams across India, Dubai, Singapore and Spain.
  • Architected enterprise systems for Rosetta Stone and RPX Insight at Accenture.
3 monthsTulifo, from blank repo to production

Customer discovery · Rapid prototyping · Production rollout

02 · Agentic AI

Agents that run the whole workflow, not a chat window.

An agent earns its place when it carries a real process end to end — and hands the decision to a person.

  • Xiya runs a fully autonomous interview: it asks its own follow-ups, scores against a rubric and explains its recommendation.
  • Orchestration in Python and FastAPI on the Claude API and Agent SDK.
  • Workflow automation that turns multi-step manual processes into agent runs.
End to endXiya runs the full interview; people make the call

Claude Agent SDK · LiveKit · Workflow automation

03 · Retrieval-augmented generation

Answers grounded in your data, not the model's memory.

Retrieval is what makes a model useful on a company's own information — so it gets the same care as the model itself.

  • RAG over candidate and job data powers Tulifo's matching.
  • The verified pool matches pre-interviewed candidates to employers whose role fits.
GroundedTulifo matching runs on real candidate and job data

RAG pipelines · PostgreSQL · Redis

04 · LLMs in production

Models chosen, trained and shipped like any other system.

Model choice, training and cost are engineering decisions — owned, measured and documented.

  • Own model selection at Tulifo: in-house Xiya and the open Chittu models alongside Anthropic and OpenAI.
  • Fine-tuned Chittu Guard from Qwen3-4B with MLX, and trained Chittu Story from scratch on 2.33M stories.
  • Every Chittu release ships with a model card listing its evaluations and known failures.
2 open modelsChittu Guard 4B and Chittu Story 26M on Hugging Face

Hugging Face · Fine-tuning · MLX · Model cards

03

Now: CTO at Tulifo.

Tulifo trains its own models to run fair, verifiable interviews for any role, in any language. I've architected and built all of it — the model work, the products and the platform underneath.

  • 2026founded · Coimbatore and Bengaluru
  • 6products on one model family
  • 10interview languages
  • 3 monthsfrom blank repo to production

How an interview with Xiya works — select a stage

Candidate joins

Xiya introduces the role and explains how the session works, in the candidate's own language. Consent is recorded and versioned before anything is shared.

Any role · 10 languages · Consent first

Six products, one model family

IInterview

Fully autonomous video interviews for any role, in the candidate's language, with background verification built in.

 JJobs

Openings matched to the skills a candidate already has, with one-step apply.

 VVerified pool

Candidates who've interviewed and cleared background checks once, matched to employers whose role fits.

 RRehearsal

Practise a Google-style or Microsoft-style loop with interviewers from top tech companies.

 ppinmbo

A learning companion for kids, built on Chittu, that gives hints instead of answers.

 CChittu

Tulifo's own language model family, with open weights on Hugging Face.

Models in my stack

Tulifo · in-house

Xiya

The interview model, trained for follow-up questions, rubric scoring and fair pacing — with a focus on non-Western-English speech.

Hugging Face · open weights

Chittu Guard

A 4B guard model fine-tuned from Qwen3-4B with MLX that escalates with four tokens instead of writing crisis replies itself.

4B · 5,250 synthetic examples · Apache-2.0 ↗

Hugging Face · open weights

Chittu Story

A 26M story model trained from scratch on children's stories swept by Chittu's own guards. No real child data.

26.3M · 2.33M stories · Apache-2.0 ↗

Frontier LLMs

Claude and OpenAI

Anthropic Claude through the API and Agent SDK for reasoning and orchestration, with OpenAI models alongside — chosen per job.

04

Six countries, five teams.

  1. Six countries · 2012 — now

    Around the world

    The places I've built teams and products, from Chennai to Spain.

  2. India · 2012 — now

    Chennai

    Home base. Tulifo, Francium, Accenture and Wipro.

  3. United States · 2021 — 2024

    Watermark Insights

    Led a 45-person remote engineering organisation with 29 direct reports.

  4. China · 2021 — 2024

    Watermark Insights

    An engineering team in China, part of the same organisation.

  5. Dubai and Singapore · 2018 — 2021

    Francium

    Distributed teams across India and Dubai, and fintech and mobility client work.

  6. Spain · 2018 — 2021

    Francium

    Client and delivery teams.

Teams I've led team direct reports

  • Watermark InsightsEngineering Manager4529 direct
  • WiproSenior Technical Lead12
  • TulifoCTO10
  • FranciumSolutions Architect98 direct
  • AccentureTechnical Architect5

Clients and products along the way: Rosetta Stone, RPX Insight, MoneySmart, Spin and Base))) Harmony.

  1. 2026 — now

    Tulifo · India

    CTO

    Architected and built Tulifo end to end: Xiya the interview model, six products on one model family, and the open-weights Chittu models on Hugging Face. Own architecture, cloud, model selection, security and roadmap, and shipped the platform to production in three months.

    Python · FastAPI · Anthropic Claude · Hugging Face · LiveKit · GCP and AWS

  2. 2022 — 2024

    Watermark Insights · US, remote · with a team in China

    Engineering Manager

    Led a 45-person engineering organisation with 29 direct reports, with people in the US and China, and stayed hands-on as a full-stack contributor.

    −70%production bugs
    15% → 3%attrition
    30%of the team hired in 2 years
    95%cross-trained to full stack
  3. 2021 — 2022

    Watermark Insights · US, remote · with a team in China

    Technical Architect

    Designed the Student Learning & Licensure platform for US higher education and led its migration from a legacy monolith to microservices on AWS. Promoted to Engineering Manager.

    +40%performance at peak
    −30%security vulnerabilities
    Awardof Excellence, 2022
  4. 2018 — 2021

    Francium Technologies · India, Dubai, Singapore and Spain

    Solutions Architect

    Designed and scaled cloud-native fintech and mobility platforms for clients including MoneySmart and Spin on Kubernetes (GKE), guiding distributed teams on scalable, secure-by-design architecture.

    GCP · Kubernetes · Node.js · React · Java

  5. 2014 — 2018

    Accenture · India

    Technical Architect

    Architected enterprise-scale systems for Rosetta Stone and RPX Insight. Star Performer of the Year, 2017.

    Ruby on Rails · Ember.js · EKS · Datadog

  6. 2012 — 2014

    Wipro · India

    Senior Technical Lead

    Led technical delivery and architecture decisions for the Base))) Harmony product across a 12-person team.

    Ruby · Node.js · MongoDB · AWS

What I'm hired for

Agentic AI

LLM orchestration, RAG and real-time voice agents, taken all the way to production.

Claude · OpenAI · RAG · LiveKit

Cloud architecture

Cloud-native systems on GCP and AWS, and monolith-to-microservices migrations.

AWS SA Professional · GKE · EKS

Engineering leadership

Hiring, growing and keeping teams of up to 45 across time zones.

Strategy · Hiring · SAFe 5

Backend and APIs

Services in Python, Rails, Node.js and Java on data stores chosen for the job.

FastAPI · Rails · PostgreSQL · Redis

Frontend and mobile

React and Ember interfaces and Flutter apps, built with accessibility in mind.

React · Ember · Flutter · Accessibility

DevOps and security

Containers, CI/CD and observability, with security designed in rather than bolted on.

Docker · CI/CD · New Relic · Datadog

05

Selected work and writing.

All repositories on GitHub ↗

Writing

70+ Ruby katas and algorithm notes on my blog ↗
06

Credentials.

Certifications

  • AWS Certified Solutions Architect – Professional
  • Trusted Accessibility Developer
  • SAFe 5.1 Practitioner
  • Managerial Accelerator Program, 15Five

Awards

  • Award of Excellence, Watermark Insights
  • Star Performer of the Year, Accenture
  • National-level hackathon winner — infracity
  • HackerRank badges: Ruby, SQL, Problem solving

Education

  • General Management Programme
  • M.S. Software Systems Engineering
  • B.E. Electronics and Communication

07 — Contact

Let's build something that lasts.

Open to AI advisory, speaking and select collaborations.

Connect on LinkedIn

↑↓ move↵ selectesc close