Open to full-time roles · Noida, India

I build production systems,
and the pipelines that judge the models that write code.

Software engineer building production systems and evaluating the models that write code. 2+ years on Node, Next.js and AWS. Currently engineering LLM evaluation infrastructure for a Microsoft frontier-model program at Quess Corp.

Resume

years of experience: 2+
startups: 4
CGPA, B.Tech CSE: 9.4
public repos: 60+

Now

Engineering LLM evaluation infrastructure for a Microsoft frontier-model program at Quess Corp

$50k/yr saved

AWS cost initiative

GPT-5.2 vs internal

SxS benchmarking

10,000+ concurrent

SafeReport submissions

01 · About

Evidence over opinion. Boring infrastructure.

Software engineer with 2+ years of experience shipping full-stack products on Node, Next.js and AWS, and the last year on LLM evaluation: engineering the code-migration harness behind a Microsoft frontier-model program, designing rubrics and the platforms that score them. B.Tech CSE at Galgotias University (9.4 CGPA), graduating 2026.

How I work

Evidence over opinion

Every judgement gets a rationale. Scores without reasons are noise.

Boring infrastructure

ALB + ASG + a pipeline that deploys on push. Clever infra is a liability at 2 am.

Types at the boundary

Validate inputs once, trust them everywhere. TypeScript end to end.

Ship, then harden

Get the fuzzy version in front of people, then make it reliable and fast.

Measure before optimising

Profile first, then fix the slowest thing. Latency 1.86s to 1.2s came from numbers, not guesses.

Own the outcome

From empty repo to production and post-launch support. If it is live, it is mine to keep healthy.

Now

Engineering LLM evaluation infrastructure for a Microsoft frontier-model program at Quess Corp
Going deeper on Kubernetes and observability
Strengthening system design and distributed systems fundamentals

Facts

B.Tech, Computer Science and Engineering, Galgotias University · CGPA 9.4 / 10 · Sep 2026
Noida, India · Remote, hybrid or on-site (NCR / Bengaluru). Open to relocation.
Open to: Full-Stack Engineer, Backend Engineer, AI / LLM Evaluation Engineer, Frontend Engineer

02 · Experience

Shipped at every stage, from first commit to production.

Roles across product startups, from real-time chat and MERN products to AI evaluation infrastructure.

Nov 20232+ years · 4 companiesPresent

202420252026

04 / 04 · LLM evaluation infrastructure for a Microsoft frontier-model program

Quess Corp

Software Engineer - Gen AI & Full StackClient · Ethara AI Gurugram, India

Feb 2026 → Present

10 mocurrent

Engineered the evaluation infrastructure behind a Microsoft frontier-model program, deployed with client Ethara AI: the code-migration harness, the Docker-based execution layer, SFT data pipelines and the graders that decide what reaches post-training.

Architected and owned an end-to-end code-migration evaluation harness for a Microsoft frontier-model program, extending the RepoTransBench research benchmark into a production evaluation workflow adopted by the tasking team.
Engineered 1,200+ multi-architecture Docker images (amd64, arm64) across 16 programming languages for LLM evaluation infrastructure, parallelised builds with Docker BuildKit and published to AWS ECR for high-throughput evaluation.
Built an end-to-end Python data pipeline generating a Supervised Fine-Tuning (SFT) dataset for a frontier coding model: mined bounty-linked GitHub issues, filtered issue-resolving pull requests, extracted verified fix and test patches, and produced structured JSONL ground-truth instances across 16 languages.
Designed a registry-driven evaluation harness with per-repository build, dependency and test configurations for 50+ open-source projects (Redis, curl, systemd, Valkey, DiceDB), enabling deterministic, automated scoring of LLM-generated patches in isolated containers.
Owned grading and output verification: produced specification-compliant output bundles and agent trajectories, audited every sample end to end, and ran binary-rubric and side-by-side (SxS) evaluations against GPT-5.2 and Gemini-3 Pro, delivering structured JSON reports used in post-training analysis.

Feb 2026 – Present

03 · Work

Things that actually ship.

For each project: the problem, what I built and what changed.

Product · live

Agent Studio

A visual builder and time-travel debugger on top of an execution and evaluation harness for multi-agent workflows

Problem: When an AI agent fails at step six, you re-run all six steps and guess. Nobody can see what the state looked like at the moment it broke, and every retry costs the same model calls again.
Outcome: A full run of the demo workflow costs about $0.00007 and a fork re-runs nothing it already did. Deployed end to end on free tiers.

What I built

Execution harness on LangGraph: a save point after every step, so runs can pause for a human, resume, or fork with edited state.
One event log streamed live over WebSockets with replay on reconnect. Canvas, console, timeline and cost all read from it.
Evaluation harness: score outputs with rules or a judge model, see which step failed on the canvas, pick the cheapest passing model.
MCP tools plug in as nodes, user Python runs in a locked-down Docker sandbox, and the public API is rate-limited with a daily budget.
Free-tier production: Postgres migrations, CI with 86 tests plus end-to-end, caching that cut page loads from 1.4 s to 0.25 s.

Next.jsReact FlowTypeScriptFastAPILangGraphPostgreSQLGroqMCPDocker

Live Code

Product

SafeReport

AI-powered anonymous crime reporting platform

Problem: Reporting an incident is slow, intimidating and rarely anonymous. People give up before the report is filed.
Outcome: A complete report in under a minute, tracked until it is resolved, with the reporter's identity never stored.

Next.jsTypeScriptPrismaNeonDB (Postgres)Gemini AI+3

Live Code

Open source

Production-grade Backend

Typed Node/Express API with a real CI/CD path to AWS

Problem: Most starter backends stop at `npm run dev`. This one ships.
Outcome: Push to main, live on EC2 a few minutes later. Used as the base for my own services.

Node.jsExpressTypeScriptMongoDBDocker+2

Code Template

Open source · most-starred

AWS Deployment with CI/CD Guide

From SSH + PM2 + Nginx to Docker + ECR + GitHub Actions

Problem: Deployment tutorials either stop at 'it works on EC2' or assume you already run Kubernetes.
Outcome: A guide other developers actually use; my most-starred repository.

AWS EC2NginxPM2DockerAWS ECR+1

Code

04 · Try it

Grade a PR the way I do.

At Quess Corp I score coding-model outputs with task-specific binary rubrics, weighted toward correctness, then compare side by side. Here is a tiny real-style example. Grade it, then see my scoring and why.

TaskReturn the last N orders for a user, newest first. Default N to 10.

 // orders.ts
−export async function recentOrders(userId: string, n: number) {
−  const orders = await db.order.findMany({ where: { userId } });
−  return orders.slice(0, n);
+export async function recentOrders(userId: string, n = 10) {
+  const orders = await db.order.findMany({
+    where: { userId },
+    orderBy: { createdAt: "asc" },
+    take: n,
+  });
+  return orders;
 }

Model's own summary of the change

“Moved sorting and limiting into the database query, added a default limit of 10 and newest-first ordering.”

Binary rubric · 100 pts

50% correctness-weighted

Correctness · 30 pts
Orders come back newest first.
Correctness · 20 pts
Limiting happens in the database, not in memory.
Code quality · 15 pts
Default parameter and query shape are idiomatic Prisma.
Behavioural reasoning · 15 pts
Handles edge input (n <= 0, very large n) sensibly.
Summary accuracy · 20 pts
The model's summary matches what the diff actually does.

05 · Skills

The toolkit, grouped by where it is used.

Production experience across the stack, with a current focus on evaluation infrastructure and platform reliability.

Programming Languages

TypeScriptJavaScriptPythonSQLC++

Application Engineering

ReactNext.jsReact NativeAngularTailwind CSSFramer Motion

Backend Systems & Data

Node.jsExpress.jsPostgreSQLMongoDBPrismaRedisGraphQLREST APIs

Cloud Infrastructure & Delivery

AWS (EC2, S3, CloudFront, Lambda)DockerKubernetesGitHub ActionsCI/CD pipelinesServerless

LLM Evaluation & Applied AI

Rubric designSxS evaluationModel benchmarkingMulti-turn scoringLLM APIsPrompt engineering

Architecture & Reliability

MicroservicesEvent-driven systemsDistributed systems designObservabilityZero-downtime deployments

06 · People

What people I've worked with actually wrote.

From founders, leads and engineers I have worked with. Each one links to LinkedIn.

I had the pleasure of working with him on multiple projects, where he showcased a strong understanding of React.js and Next.js concepts while delivering high-quality code. He consistently demonstrated a willingness to learn and grow, and his enthusiasm for tackling new challenges was truly inspiring. Highly recommended!

Abhijit Agarwal

Founder, Xipper

I had the privilege of working alongside Prince on multiple industrial projects, and his talent and work ethic truly stood out. His professionalism and attention to detail were remarkable. I wholeheartedly endorse Prince for any opportunity that requires a highly skilled and dedicated individual.

Sathya Sachi Paira

CEO, SecWebXperts

Had the pleasure of working with Prince during his internship at Xipper. Quick to learn, made essential contribution with developing the MVP, and showed strong initiative throughout. A reliable and thoughtful developer anyday!

Shyan Roy Choudhury

SDE-2, Publiq Studio

A talented React frontend developer with an exceptional eye for responsiveness. His proficiency in Node.js further enhances his capabilities, enabling him to build robust and scalable applications. Attention to detail, problem-solving skills and commitment to delivering outstanding results.

Ayush Kumar Tiwari

Software Engineer, D4 Community

He did a great job during his time with us, especially in building clean and user-friendly web screens for our project at Torus. Prince is hardworking, quick to learn, and always delivered his tasks on time.

Deepanshu Kamboj

Full Stack Developer, Devtown

He's quick to understand requirements, writes clean and efficient code, and always brings a positive attitude to the team. Whether it's tackling tricky UI bugs or building smooth user interfaces, Prince handles it all with focus and care.

Harshit Bafna

Full Stack Developer, Akal Infotech

07 · Proof

Public track record.

GitHub activity, competitive programming profiles and certifications.

github.com/HardCoder404

public repos

…

stars earned

…

followers

…

Problem solving

1800+

LeetCode contest rating · 3-star

700+

DSA problems solved

1200+

Codeforces rating

GSSoC'24

open-source contributor

LeetCode badges

Streak and daily-challenge badges earned on leetcode.com/u/HardCoder404.

14 badges

AWS Academy Graduate · Cloud Foundations

AWS Academy Graduate · Cloud Architecting

C++ · Certificate of completion

08 · Contact

Hiring, or just curious?

I reply to every real message, usually within a day. Email is fastest.

+91 82183 28333

Open to full-time roles

Full-Stack Engineer · Backend Engineer · AI / LLM Evaluation Engineer · Frontend Engineer

Remote, hybrid or on-site (NCR / Bengaluru). Open to relocation.

Scan to save my contact

Opens as a contact card on your phone: name, number, email and this site.

Book a 15-minute callSkip the form. Pick a slot that works for you, the Google Meet link arrives with the invite.pick a slot →