Senior Software Engineer · Microsoft Azure · Bengaluru

I build AI systems that hold up in production, and I want to point them at problems that matter.

For more than eight years I've designed large distributed systems at Microsoft, Walmart and Ola. Today I build production AI systems at Microsoft, using LLMs, RAG and multi-agent workflows. I'm especially drawn to problems where AI meets public institutions in India.

Get in touch LinkedIn GitHub
Portrait of Rajesh Sahu, smiling, against a dark studio backdrop
8+ yearsbuilding production systems at Microsoft, Walmart, Ola and Accenture
Pinnacle 2026Microsoft Pinnacle Award, for impact with applied AI
25+ buildsproducts I've built on my own time since 2017, from Android apps to multi-agent AI
M.Tech, AIBITS Pilani, plus a Springer best paper on deep learning for security
Built at Microsoft Walmart Ola Accenture
Focus

What I care about

Most people who understand how frontier AI systems are built have never had to make something work inside a court, a district office or a public program. Most people who understand those institutions have never shipped a production AI system. That gap is where I find the most interesting problems.

Three things guide what I work on:

01

Technical depth

Staying at the frontier of applied AI: multi-agent systems, evaluation, grounding and reliability at cloud scale.

02

Institutional understanding

How India's administrative and legal systems actually work at scale, well enough to build for them, not just critique them.

03

Useful over impressive

AI that holds up when real people depend on it, in courts, offices and small businesses, not just in demos.

Now

What I'm focused on

  • Applied AI at Microsoft. Building production systems with LLMs, RAG and agents, and the evaluation that makes them trustworthy.
  • Building AI for legal practice in Odisha. A drafting assistant for chambers at the Orissa High Court and district courts, grounded in the case record and with no invented citations.
  • Writing in public. Essays on what it actually takes to make LLM systems reliable, from evals to grounding to failure modes. They'll be published here first.
  • Meeting people. Researchers, engineers building AI in India, and people who work inside public systems.

Updated October 2026

Experience

Where I've done the work

Microsoft logoMicrosoftSenior Software Engineer
Oct 2024 – present

Applied AI for cloud software quality

  • Came up with and lead an AI-native platform that uses ML models, LLMs, RAG and multi-agent workflows to automate software quality engineering at cloud scale. I own it end to end, from architecture to production operations.
  • Built evaluation and observability for AI-generated output, so it can be trusted in production decisions.
  • Lead design and code reviews, and help engineers build AI systems that are reliable as well as impressive.
Pinnacle Award 2026AI Core Demo Day 2026 winnerHackathon 2025 winner

Reliability engineering

  • Built monitoring, anomaly detection and automated release quality checks for large-scale cloud services, catching regressions before they reach production.
  • Worked with cross-functional teams on release criteria and self-healing mechanisms to improve service availability.
Walmart logoWalmart Global TechSenior Software Engineer
Nov 2023 – Oct 2024
Software Developer III
Dec 2021 – Sep 2023

High-scale messaging platform, built from scratch

  • Led the end-to-end architecture of a high-scale, event-driven messaging platform for customer communications, and its migration to production with zero downtime.
  • Scaled it more than tenfold while meeting strict latency targets for priority messages.
  • Applied resilience patterns on Kafka: retries, circuit breakers, bulkheads and rate limiting.
  • Led and mentored a team of 7 engineers, and owned production reliability, incident response and root-cause analysis.
Walmart Impact Award 2022
OlaSDE 2, Security Platform
Aug 2021 – Dec 2021
  • Built high-availability infrastructure and APIs for identity and access management, and fixed scalability bottlenecks in production.
  • Built an audit-trail service that streams access events to Kafka for real-time compliance monitoring.
Accenture logoAccentureInnovation Hub
Aug 2018 – Jul 2021
Associate Application Developer → Application Development Senior Analyst
  • Machine learning: built an ML-driven pricing engine with Java microservices and an analytics dashboard, deployed on AWS.
  • Conversational AI: built services and webhooks for a Dialogflow CX healthcare support assistant.
  • IoT and mobile: built real-time indoor navigation with IoT beacons, and Android apps.
Innovation Contest finalist, from 400k+ACE Award 2021Spark Award 2020
Independent builds

Things I build on my own time

I learn by shipping. I've built more than 25 products on my own time since college, from Android apps and marketplaces to multi-agent AI systems. Here are the ones I'm working on now, followed by the archive.

Legal AI · Odisha courts Working prototype

Chamber Drafting Assistant

ClaudeRAGFastAPIPostgreSQLReact

An AI drafting workspace for law chambers. It reviews every paragraph of a bail application or plaint against the case record and a risk rubric, translates scanned Odia documents, and runs mock hearings against simulated benches to prepare counsel.

Design rule: no invented citations. The model can only cite judgments that were retrieved, every quote is checked against the source page, and severity comes from the rubric, not the model.

Multi-agent systems Personal use

LifeOS

Multi-agentPythonClaude

A self-evolving team of agents that keeps an evidence-based model of one person's life. A Profiler, domain specialists, a Strategist and a Critic, and a Forecaster turn daily check-ins into a plan. An Agent Architect hires, refines and retires specialists based on how useful they turn out to be.

Lesson: agent teams improve when usefulness is measured every run and the system can change its own structure, within strict limits.

Local commerce Deployed on Firebase

Sikka

React NativeFirebaseVertex AI

A QR-code loyalty program for neighbourhood stores. Customers scan to earn points, managers approve in real time, and stores post offers targeted by location. Ad images are generated from a one-line prompt.

Why: small Indian merchants deserve the retention tools large chains take for granted.

Food · D2C Early milestone

Chakki

ReactCloud FunctionsAI recommendations

Made-to-order atta. Customers design their own flour blend, see the nutrition update live, and get it freshly ground at a mill and delivered. It has separate apps for customers and for the mill floor.

Why: to bring personalisation to a staple that every Indian kitchen uses daily.

Quant · Indian markets Design + build

NSE Signal Stack

PythonML rankingBacktestingKite Connect

A human-in-the-loop swing-trading signal engine for NSE and BSE. It covers a data layer, features, a model roadmap, honest backtesting, risk management, and compliance under Indian market rules.

Design rule: the model's job is to rank and size opportunities consistently, not to predict the market. The edge comes from discipline, coverage and execution.

AI · Fashion In development

Fashionista

FlutterFirebaseGemini

An AI-powered styling platform that connects shoppers, stylists and e-commerce vendors, built as one iOS and Android codebase on a fully serverless Firebase backend.

Why: personal styling shouldn't be a luxury. AI can bring a stylist's judgement to anyone with a phone.

Giving back · NinjaCodeClub, a free education initiative

Helping engineers break into big tech

NinjaCodeClub is a free initiative where I share what I've learned about data structures, system design and interviewing, to help more engineers reach top product companies. It's free, as a way of giving back.

Visit NinjaCodeClub →

More things I've built since 2017

All repositories on GitHub →
2023 – 26Honour Our Loved OnesA digital memorial for celebrating lives well lived: share memories and light virtual candles together. It began in 2023 as HOLO ("Let's remember together").TypeScript · Gemini · Firebase
2020 – 25SureSortedLiveA voice-first app for finding local help (electricians, plumbers, carpenters, tutors), with a separate partner app for service providers. Rebuilt in 2025 with Capacitor for iOS and Android.TypeScript · Java · Firebase
2022 – 25Dealer network appsFlutter apps for a business client's dealer network, on phone and tablet, built with a clean-architecture layout and maintained from 2022 to 2025.Flutter · Dart
2024Crowdfunding platformA full-stack crowdfunding app with a Spring Boot API and a React frontend (frontend).Java · Spring Boot · React
2022ReferTubeA job-referral marketplace: "Unlock the power of referrals and land your dream job with ease." It followed the 2019 GetReferral portal.React · Firebase
2022KidsClub & FamilyTreeTwo Flutter apps: a children's club app and an interactive family-tree builder, both with feature-first architecture.Flutter · Dart
2021MatchedA social app that matches people by personality and interests into sessions, interest groups and help circles.Java · Android
2020JukeBoxA Node.js and Express backend for a music library and jukebox service.JavaScript · Node
2019 – 20JustFashionistaA headless e-commerce storefront and dashboard built on Saleor (GraphQL, Django, React). It was the first version of the Fashionista idea.Python · GraphQL
2019Blaze, CWS & SlogNative Android apps from my early mobile work.Java · Android
2018 – 19DeStreetsA platform for street talent, with online "Battle of the Streets" competitions, registrations and artist profiles, on web and Android.PHP · Java · Android
2017GreenshelfA marketplace for buying and selling used books, built in college.PHP · MySQL
2017RedeyeGeneric commercial billing software for small businesses.Java · SQLite
2017Part-time job schedulerSchedules student part-time workers around their college timetables.C++
2017Rank Your PeopleA live-ranking web app for fun peer rankings among classmates.PHP · AJAX
Writing

Notes from building systems at scale

I write about the engineering behind high-throughput, event-driven systems, and about where AI can actually help people. More essays on LLM reliability are coming.

All posts on Medium →

Recognition
  • 2026Microsoft Pinnacle AwardRecognition for impact with applied AI
  • 2026Winner, Microsoft AI Core Demo DayApplied generative AI
  • 2025Winner, Microsoft HackathonApplied AI
  • 2022Walmart Impact AwardA product taken from idea to production
  • 2021Accenture ACE AwardFuture-ready skills category, Accenture ATCI
  • 2020Finalist, Accenture Technology Innovation ContestSelected from more than 400,000 applicants
  • 2020Accenture Spark AwardRecognised as “The Learner”, Accenture ATCI
  • 2019Best Research Paper, ICACIE (Springer)Application of deep learning for database intrusion detection
Education
  • 2025M.Tech, Artificial IntelligenceBITS Pilani · 2023 – 2025
  • 2018B.Tech, Computer Science & EngineeringVSSUT Burla, Odisha · CGPA 8.45
Toolkit

Generative AI

LLMs, RAG, multi-agent systems, MCP, evals, prompt engineering, AI-native application design

ML & deep learning

Supervised and unsupervised learning, neural networks, model training and evaluation, anomaly detection, ML-driven pricing and decision systems

NLP & conversational AI

NLP, Dialogflow CX, intent-based chatbots

Systems

Java, Spring Boot, TypeScript, Python, .NET, Kafka, gRPC, Dropwizard, Dagger, microservices

Cloud & DevOps

Azure (Functions, OpenAI, Service Bus, Cosmos DB), AWS, GCP, Firebase, Docker, Docker Swarm, Jenkins, CI/CD, telemetry and monitoring

Data

Spark, Hadoop, Hive, Dataproc, Airflow, MySQL, Cassandra, Redis, MongoDB, BigQuery

Contact

If you're working on any of this, I'd like to talk.

Builders and founders

Especially if you're building AI for India, or need someone who has taken agentic systems from prototype to production.

Researchers

Especially on evaluation, grounding and reliability of LLM systems. I'm always happy to compare notes.

People in public systems

Courts, administration, public programs: if you've seen where technology adoption stalls, I want to learn from you.

rksahu.1214@gmail.com LinkedIn GitHub Medium