[ SOFTWARE ENGINEER · 10 YRS ] Architecting agentic systems & the platforms underneath them.

KSHITIZ BHARA.DWAJ

Role
▸ Software Engineer  ·  Agentic AI  ·  Systems Architect
Based
Noida (NCR), India  ·  Working globally / async
Now
Engineer @ Thales  ·  Building Atomic on the side
Index
01 About  · 02 Expertise  · 03 Work  · 04 Trajectory  · 05 Contact
28.5°N · 77.3°E Portrait of Kshitiz Bharadwaj FIG.01 — OPERATOR
// 01ABOUT

A decade of building.
From sketch to scale.

27, ten years in. By day, core engineering & DevOps at Thales. By night and weekend — where most of my agentic AI, system-design and solutions-architecture work lives — the framework I'm building (Atomic) and a long arc of side projects across AI, web, and ventures.

a brief on the operator

I'm a software engineer, 27, with about a decade of building behind me — starting in 2016 with freelance graphic and UI/UX work, then a couple of ventures, then years of contract development, and now full-time architecture and AI work at Thales.

Two of my own startups closed along the way: Futura (AR/VR for enterprise) and Lumis (e-commerce). Both failed. Both taught me more than they cost. Between them: contract development at Prime Gurukul on ed-tech, plus two stretches of independent SaaS work for various clients.

what I'm building

By day, at Thales: Java / Spring Boot core development on license-management platforms, plus Selenium / Playwright automation, CI/CD & GitOps on Kubernetes & GCP, observability with Datadog, and secret management with HashiCorp Vault. The unglamorous engineering that keeps a mature platform alive.

By night and weekend, the other half — Atomic, a model-agnostic harness framework for production AI. Past freelance projects in system design, solutions architecture, and the integration work I've done for small businesses over the years. Currently turning over world models, AR/VR's second wave, and what system design looks like when AI eats more of the stack.

// 02EXPERTISE

Four lanes,
one operator.

I don't believe in over-specializing. The interesting work sits in the seams — between research and product, between agent and UI, between a sketch and a deployable system.

01 / 04
{ }

Agentic AI &
Harness Engineering

Frameworks like Atomic that abstract LLM providers, expose unified tool-use, and turn "an LLM in a loop" into something a team can actually deploy.

SPECIALTY
02 / 04
▤

Systems
Architecture

Designing systems that hold up. One client system I shipped scaled to ~10 lakh concurrent users in a month — backend, queues, graceful degradation, observability.

CORE
03 / 04
☰

Cloud
& DevOps

Production infrastructure across Kubernetes, GCP, Datadog, and HashiCorp Vault. CI/CD pipelines, GitOps workflows, Selenium & Playwright automation — the parts that decide whether 3am pages happen.

INFRA
04 / 04
◀

Product
Engineering

Java / Spring Boot backends, React frontends, the connective tissue in between. Ten years of shipping for real users — from solo SMB tools up to enterprise platforms.

DELIVERY
Java Spring Boot React Kubernetes Agentic AI GCP Docker Datadog HashiCorp Vault GitOps CI/CD Selenium Playwright Microservices REST APIs PostgreSQL System Architecture Observability LLM Orchestration RAG Tool-use Protocols AI / ML AR / VR World Models TypeScript
// 03SELECTED WORK

Five systems,
five kinds of stakes.

Frameworks I'm building, systems I've shipped at scale, AI products in regulated domains, and the long tail of integration work I've done for small businesses on the side.

/01

Atomic

A model-agnostic harness framework for production AI — unified interface across LLM providers, runtime tooling, instrumentation. The layer that lets teams ship AI features without locking into a single provider's roadmap.

Architect / Author 2025 — Active Harness Eng.
＋

Atomic is what I'd reach for if I had to ship ten AI features tomorrow. It treats LLM providers as interchangeable backends, exposes a consistent tool-use surface, and gives engineers the observability hooks they need to debug a model the way they'd debug any other service.

The opinionated bit: it's biased toward shipping. No magic, no surprise abstractions — primitives that compose, with the seams visible to the engineer using them.

Role
Architect & author
Domain
Harness Engineering · Production AI
Status
In active development
// stack
  • Java / Spring
  • LLM Runtime
  • Provider Abstractions
  • Tool-use Protocol
  • Observability Hooks
/02

DBA-x

A framework for talking to SQL databases in natural language — letting customers query their data without ever writing a join. An LLM in front of a real database, with the rails it needs to stay honest.

Builder / Designer 2026 — Active NL Interfaces
＋

DBA-x puts an LLM in front of a real database with schema awareness and query inspection — the user sees exactly what's about to run before it runs. Read/write separation, ambiguity surfaced back to the user, destructive operations gated behind explicit consent.

The hard part was never "generate SQL." It was making the model know what it doesn't know — and learning to refuse, gracefully, when it didn't.

Role
End-to-end builder
Domain
NL → Data · Customer-facing
Outcome
Production pilot
// stack
  • Python / Java
  • LLM Tool-use
  • Postgres / MySQL
  • Schema Indexing
  • React Frontend
/03

At Scale [NDA]

Designed and shipped a system that scaled to 5–10 lakh (~500K–1M) concurrent users within a month of launch. Backend, queues, graceful degradation, observability across the path.

System Architect NDA'd Scale
＋

The lesson wasn't "how to scale" — it was how to architect so that scaling doesn't require a rewrite. The right primitives upfront meant the system absorbed a 1000× ramp in user count without any of the panic refactors that usually accompany it.

The product and client are NDA-bound. The patterns and trade-offs that made it work are what stuck with me.

Role
Architect & lead engineer
Domain
Consumer scale · NDA
Outcome
Shipped & sustained
// stack
  • Java / Spring
  • Kubernetes / GCP
  • Queueing
  • Datadog
  • Graceful Degradation
/04

AI for Health

A suite of AI applications for the healthcare industry — including a domain-trained chatbot for clinical workflows. Hard guardrails, audit-friendly outputs, deep schema awareness, and a long view on what the system should refuse to do.

Architect / Engineer 2024 Healthcare AI
＋

Healthcare AI lives or dies on what the model refuses to do. Most of the work was the boundary — making the system honest about uncertainty, surfacing the right paper trail to the clinician, and refusing to wing it on edge cases.

My role was architecture and the chatbot's reasoning layer; the rest shipped with a small, focused team.

Role
Architect & engineer
Domain
Healthcare · Regulated
Outcome
Deployed in production
// stack
  • Python / Java
  • LLM Tool-use
  • Domain Guardrails
  • Audit Trails
  • React Frontend
/05

SMB Connectors

Integration work for small businesses over the years — connecting their tools, automating the boring parts, exposing data they couldn't otherwise see. Low-overhead systems, real margin impact, ship-and-disappear.

Solo / Engineer Various years SMB
＋

The work most engineers skip and most SMBs can't afford on a salary. Each project was small, opinionated, and shipped. The thread across all of them: pick the smallest system that solves the problem, then disappear.

Role
Solo engineer
Domain
SMB · Integration · Automation
Status
Past projects, various years
// stack
  • Java / Node
  • REST & Webhooks
  • Postgres
  • Lightweight Auth
  • Cheap Hosting
// 04TRAJECTORY

A decade.
Two ventures.
Two companies. One arc.

From freelance graphic and UI/UX work in 2016, through two startups that didn't make it, into the architecture and AI work I do today.

2022 — now  ·  Enterprise (day)

Software Engineer

Thales / License Management Solutions

Started as an intern in Feb 2022, transitioned to full-time. Java / Spring Boot core development on license-management platforms, Selenium & Playwright automation, CI/CD and GitOps on Kubernetes & GCP, with Datadog observability and HashiCorp Vault for secret management. The agentic AI, framework, and architecture work happens entirely outside this role — on the side.

2020 — 2022  ·  Independent + Contract

Contract Developer & Freelance Engineer

Prime Gurukul / Ed-Tech + Multi-client SaaS

Contractual development at Prime Gurukul on their ed-tech web application (1–1.5 years), alongside independent SaaS projects for various clients. Where I learned to be a one-person engineering org — design, deliver, deploy, support.

2020  ·  Closed

Founder — Lumis

E-commerce brand / Dropshipping

A brand experiment in the dropshipping margin model. Ran it for ~6 months, then wound it down — killed it faster than it could die slowly. The post-mortem taught me more about unit economics and retention than two years of theory.

2019  ·  Closed

Founder — Futura

AR/VR Solutions / Enterprise & SMB

My first venture — AR/VR for enterprise and small business. Closed. Market timing wasn't there, but the architecture lessons (and the immersive-interface curiosity) still inform my work today. Every world-models thought-experiment I have traces back to this.

2016 — 2018  ·  Independent

Freelance Engineer & Designer

Self-directed / Multi-client

Side projects, graphic design, UI/UX work. The start of a habit I still keep — shipping for real users, fast, with the seams visible. Where the design instinct that still shows up in my engineering came from. A decade in: two failed startups, two companies, dozens of clients, and one ongoing belief — software is most interesting when it's quietly handing power back to the people using it.

// 05CONTACT

Say hello.

LinkedIn / kshitiz-bharadwaj Back to top ↑
Connect on LinkedIn ⟶