Available to start now · Stamp 1G · Dublin, Ireland

Sandeep
Konduru Raju.

Backend engineering. Applied AI.

I spent 4 years at Manhattan Associates building the Java and Spring Boot services behind a warehouse platform that handles 10M+ transactions a day, and I was on call when it broke. Now I'm finishing an MSc in Artificial Intelligence at NCI Dublin and shipping AI agents that run live, tested by their own evals.

Sandeep Konduru Raju

4 yrs at Manhattan Associates

10M+

transactions a day

99.5%

platform availability

15K+

requests a second

100+

production incidents fixed

Now

  • Finishing my MSc thesis on ROUGE vs BERTScore
  • Running Dublin Transit AI live on Azure and watching its eval on every change
  • Learning and stress-testing every design decision in my incident platform (v1.0.0)

01 / Experience

Built for real operations.

Sep 2021 — Dec 2025

Bengaluru, India

15 retail clients, Europe and US

Software Engineer

Manhattan Associates

Built and ran Java microservices and REST APIs for a warehouse management platform (inventory, orders, items, transportation).

  • Kept the platform handling 10M+ transactions a day at 99.5% availability.
  • Raised throughput to 15K+ requests a second and cut latency by 45% by rewriting the slowest code paths.
  • Built the RabbitMQ integration between the warehouse system and automated equipment for Crocs, keeping stock counts right in real time.
  • Designed the REST APIs behind 50+ customer integrations.
  • Fixed 100+ production incidents on call, using tracing and logs, and wrote postmortems so they did not come back.
  • Added JUnit and Mockito tests that cut regression defects by 40%, and mentored 3 junior engineers.
  • Helped build MAKI, an internal multi-agent coding tool on LangChain and Gemini, and built its evaluation harness. It caught made-up config names before production and cut turnaround from 24 to 12 hours.
  • Delivered 4 go-lives for enterprise retail clients.
Rising Star · Q2 2022Star of the Quarter · Q1 2023Spirit Award · Q3 2024

02 / Selected work

Small systems. Clear decisions.

Two flagship projects with live demos, then smaller experiments with code on GitHub. In the diagrams, solid boxes are built and dashed boxes are next.

AI investigation of a payment incident: high confidence root cause with numbered evidence citations
Live · v1.0.0

Incident root cause assistant

AI Incident Investigation Platform

Helps engineers find the root cause of a production incident faster. A Spring Boot API sends events through a transactional outbox to Kafka; a Python LangGraph worker runs hybrid RAG with Claude and returns a cited root cause, or abstains when evidence is weak.

Spring Boot APIOutbox + KafkaLangGraph + RAGReact UI

Hardest part: Making it abstain when the evidence is weak instead of guessing, while every cited source stays valid.

On a 28-case eval: 89% root cause accuracy, 100% valid citations. Redis cache cut read p95 from 1,042 ms to 44 ms.

  • Java 17
  • Spring Boot
  • Kafka
  • Python
  • LangGraph
  • pgvector
  • Redis
  • React
  • Kubernetes
Chat answer to: is the 39A running late right now? Live delays from 40 trips, with route check and live delay source chips
Live · v1.0

AI assistant on live bus data

Dublin Transit AI

Answers Dublin bus questions in plain English, like "is the 39A running late?" or "how reliable was route 15 this week?". A poller ingests the NTA GTFS-Realtime feed into Postgres every minute, a FastAPI service joins it to the national timetable, and a LangGraph agent with Claude answers only from those tools.

NTA live feedPoller + PostgresFastAPI toolsLangGraph agentChat UI

Hardest part: Live data silently drifted until only 13.7% of trips matched the timetable. A join-health check caught it, and a version-gated refresh brought it back to 99.9%.

A 100-question eval over frozen data with a 100% safety gate scores 100/100, and a sample runs in GitHub Actions on every agent change. It refuses routes that do not exist, like the retired 46A. Upserts plus a daily rollup cut stored rows 25x. Live on Azure behind Caddy HTTPS, rate limited.

  • Python
  • FastAPI
  • LangGraph
  • Claude
  • PostgreSQL
  • Next.js
  • Docker
  • Azure

Smaller experiments

Live on GitHub

Prompt input checks

SecureAgent

A small API that checks prompts before they reach a model. Three rule based checks vote on whether a prompt is safe, and high risk patterns are always blocked.

Prompt3 rule checksVoteAllow / block

Has JSON logging, a health endpoint and ten tests. It uses rules, not LLM classifiers.

  • Python
  • FastAPI
  • Pydantic
  • pytest
Live on GitHub · MSc

MSc / AI evaluation

Prompt injection detection

Compared a single model, a vote between three LLMs, and a sanitization step on 100 prompts, half safe and half attacks.

Measured precision, recall and F1. One finding: sanitizing a prompt can hide the signs of an attack. A separate experiment from SecureAgent.

  • Python
  • Groq API
  • LLM evaluation
  • SQLite

03 / Toolkit

What I use, and where.

  • Used in production
  • Used in my projects

Languages

  • Java 17
  • Python
  • Go
  • SQL
  • JavaScript
  • Bash
  • TypeScript

Backend

  • Spring Boot
  • Hibernate / JPA
  • FastAPI
  • REST APIs
  • Microservices
  • Event-driven design
  • OAuth2
  • JWT

Frontend

  • React
  • JavaScript
  • HTML / CSS
  • Next.js

Data and messaging

  • MySQL
  • PostgreSQL
  • Redis
  • Kafka
  • RabbitMQ
  • Flyway

Cloud and delivery

  • AWS
  • Azure
  • GCP
  • Docker
  • Kubernetes
  • Terraform
  • Jenkins
  • GitHub Actions
  • Linux

Reliability (SRE)

  • On-call
  • Incident response
  • Root cause analysis
  • Postmortems
  • Alerting
  • OpenTelemetry
  • Prometheus
  • Grafana

AI and ML

  • LLMs
  • LangChain
  • LangGraph
  • RAG
  • pgvector
  • MCP
  • Agent evaluation
  • NLP
  • Transformers

Foundations

  • Data structures and algorithms
  • System design
  • Distributed systems
  • OOP and design patterns
  • Testing (JUnit, Mockito)
  • Agile / Scrum

04 / Education

Learning with purpose.

MSc Artificial Intelligence

National College of Ireland, Dublin

Jan 2026 — Dec 2026 (expected)

Thesis in progress: how reliably ROUGE and BERTScore judge AI summaries of scientific papers.

B.E. Computer Science

Sir M Visvesvaraya Institute of Technology, Bengaluru

2017 — 2021 · CGPA 7.97/10

Let’s connect

Have a role or a problem
worth exploring?

Open to backend, full stack, SRE and AI engineering roles in Dublin. I hold a Stamp 1G and can start now.