Prabhpahul Singh

Sheet 00 · Software Engineer

Prabhpahul
Singh

Software Engineer: Backend, Distributed Systems & AI

Software Engineer with 6+ years of experience building scalable backend and full-stack applications across healthcare, insurance, travel and payments. I work in Java, Spring Boot, Python, React.js and Kafka on cloud platforms, and I ship production AI solutions using LLMs, RAG and agentic workflows.

  • 80%faster vendor onboarding at Eli Lilly
  • 20M+insurance records ingested per day
  • 200KAskX clinical-trial queries per day
  • $2Min customer refunds found by RefundGuard

01 · About

Backend first, with AI in the loop.

The path so far runs from payment transactions at Paytm, through travel-marketplace inventory at Isango and cloud healthcare services at Blue Cross Blue Shield, to insurance and clinical AI at Eli Lilly. Each step added more production load and more scrutiny on the systems underneath.

I care about reliability, performance and maintainable software that creates measurable business value. My side projects follow the same thread: storage engines, durable runtimes and supply-chain checks, each built to show how it behaves when something fails.

02 · Experience

Four companies, one thread: systems that must not fail quietly.

Newest first. Each entry lists the work I am proudest of and the stack behind it.

  1. Mar 2025 – Present

    USA

    Eli Lilly Software Engineer

    Build data platforms, backend services and production AI workflows for insurance and clinical operations, from serverless ingestion to multi-agent RAG systems.

    • Built a serverless ingestion platform on AWS S3, Lambda, Kafka and Python that replaced fragmented vendor onboarding, cutting onboarding time by 80% while processing 20M+ insurance records per day.
    • Developed RefundGuard, a Python/FastAPI audit service that analyzed 500K+ insurance records to find overcharges, enabling $2M in customer refunds.
    • Built AskX, a multi-agent platform (FastAPI, RAG, SQL agents, classifier routing, fallback orchestrator) for clinical-trial queries: 200K queries per day.
    • Executed a zero-downtime migration of the configuration database from Aurora MySQL v2 to v3 using blue-green deployment, with cutover in under 10 seconds and zero data loss.
    • Python
    • FastAPI
    • Java
    • Kotlin
    • Spring Boot
    • AWS Lambda
    • Kafka
    • RAG
    • LangGraph
    • OpenAI APIs
    • Aurora MySQL
    • React
  2. Sep 2021 – Mar 2025

     

    Blue Cross Blue Shield Software Engineer

    Built cloud-native healthcare services and AI-assisted applications, with a focus on reliability, observability and secure service-to-service communication.

    • Architected scalable backend services in Java and Spring Boot for cloud-native healthcare workflows, improving modularity, reliability and maintainability.
    • Built centralized logging and monitoring by integrating Splunk with custom Spring Boot actuators, decreasing mean time to resolution for production incidents by 25%.
    • Secured service-to-service communication across internal microservices with OAuth 2.0 and JWT authorization.
    • Java
    • Spring Boot
    • Python
    • FastAPI
    • React.js
    • TypeScript
    • Docker
    • Kubernetes
    • Splunk
    • OAuth 2.0
    • JWT
    • PyTorch
  3. Jul 2018 – May 2019

    India

    Isango Software Engineer

    Worked on backend services for a travel marketplace, keeping tour pricing and availability current across Australian destinations.

    • Developed Java and Spring Boot services to automate activity-data ingestion and synchronization, keeping tour pricing and availability current across Australian destinations.
    • Implemented Kafka-based event processing with Redis and MySQL to propagate inventory changes in near real time.
    • Integrated Cloudinary as an image manipulation and CDN layer, reducing average page load times by 30%.
    • Java
    • Spring Boot
    • Kafka
    • Redis
    • MySQL
    • AWS
    • Cloudinary
    • REST APIs
  4. May 2017 – Jun 2018

    India

    Paytm Software Developer

    Developed backend payment-processing components and the APIs that connect them to customer-facing applications.

    • Developed backend payment-processing components in Java and Spring-based technologies for transaction and business workflows.
    • Engineered C++ components for performance-sensitive application processing in transaction-oriented workloads.
    • Built REST APIs and SQL-based data integrations that connect backend services to customer-facing applications.
    • Java
    • Spring
    • C++
    • REST APIs
    • SQL
    • JUnit
    • Git
    • Maven

03 · Projects

Ten working systems, each drawn as it works.

Storage engines, durable runtimes, offline sync, bitemporal pricing, supply-chain verification and security tooling. Every drawing shows the mechanism, not a screenshot.

WAL memtable group commit flush sorted tables compact crash-tested at every I/O

Storage engine

lsmkv

An LSM-tree key-value storage engine in Rust with SSTables, bloom filters, leveled compaction and MVCC snapshots.

  • CRC-32C-checked WAL with group commit: one record and one fsync per group of writers.
  • Crash-tested at every single I/O operation through a simulated filesystem.
  • Rust
  • LSM tree
  • SSTables
  • MVCC
transaction time valid time backdated correction closed, never updated

Insurance systems

ratevault

A motor insurance quote engine with bitemporal rate tables, so any past quote can be repriced to the penny using the rates in force at the time.

  • Rate rows carry valid-time and transaction-time ranges and are closed, never updated or deleted; a database trigger rejects overlapping ranges.
  • Demo: a quote of GBP 1995.19 stays provable while the backdated-correction answer is GBP 1671.64.
  • TypeScript
  • Node.js
  • SQLite
  • Triggers
DRAFT SUBMITTED REVIEW APPROVED ISSUED raw write rejected legal edges live in the database

Workflow & data integrity

permitflow

A municipal permitting workflow where a declared statechart compiles into database constraints and triggers, so illegal transitions are rejected even for raw SQL.

  • One statechart with 11 states and 15 declared transitions drives both enforcement and citizen-facing next-step text.
  • Demo swept all 95 illegal pairs: 95 rejected, 0 accepted.
  • TypeScript
  • SQLite
  • SQL triggers
  • Statecharts
source build attest verify rebuild ACCEPT reject with a reason code

Supply-chain security

provenance-gate

A SLSA provenance verifier that checks attestations against policy, then rebuilds the artifact and diffs it to prove where the binary came from.

  • Rejects with specific reason codes such as DIGEST_MISMATCH, WRONG_BUILDER, REPLAYED and NOT_REPRODUCIBLE.
  • A compromised runner with a validly signed attestation is caught only by the rebuild-and-diff step.
  • Python
  • SLSA
  • in-toto
  • Reproducible builds

Also built

picker A queue picker B queue ledger applied exception task

Offline-first sync

pickpath

Offline warehouse picking sync. Conflicting picks become supervisor exception tasks, and on 500 contested SKUs the ledger had 0 silent losses.

entry handler dynamic call vulnerable function reachable indeterminate, never safe

Security tooling

reachable

CVE triage over static call graphs. Each advisory is reachable, not reachable or indeterminate, never a silent "safe".

finished steps crash unfinished replay from history event-sourced SQLite

Agent infrastructure

replayagent

Single-node durable execution for LLM agents, in the style of Temporal. A crashed run replays its SQLite history and runs only the unfinished steps.

replica A replica B offline merge one state

Local-first apps

syncboard

Local-first collaborative kanban: offline edits queue in IndexedDB, and Yjs CRDTs resolve concurrent edits across network partitions.

certificate life rotate rotate rotate CA outage fail closed short-lived identity, exercised constantly

Security engineering

rotationchaos

Workload-identity lab issuing SPIFFE-style certificates with 60-second lifetimes. Injected failures such as CA outage and clock skew must fail closed.

spend ceiling seasonal baseline flagged hour of week

Campaign monitoring

pacingloop

Real-time campaign pacing with hour-of-week median/MAD baselines and a damped, rate-limited controller that avoids false alarms on seasonal peaks.

04 · Skills

Bill of materials.

Everything I reach for, grouped the way a system is built.

Languages

  • Java
  • Python
  • Kotlin
  • C++
  • JavaScript
  • TypeScript
  • SQL

AI & Generative AI

  • LLMs
  • RAG
  • Agentic AI
  • OpenAI APIs
  • LangGraph
  • Vector Embeddings
  • Semantic Search
  • AWS Bedrock
  • PyTorch
  • TensorFlow
  • Scikit-learn
  • MLOps

Backend & Distributed Systems

  • Spring Boot
  • FastAPI
  • Microservices
  • REST APIs
  • Event-Driven Architecture
  • Asynchronous Processing
  • Hibernate
  • Spring Data JPA

Frontend

  • React.js
  • Angular
  • HTML5
  • CSS3
  • Responsive UI

Cloud & DevOps

  • AWS
  • Amazon S3
  • AWS Lambda
  • Amazon EC2
  • Amazon EKS
  • GCP
  • Docker
  • Kubernetes
  • Jenkins
  • CI/CD

Data & Messaging

  • Apache Kafka
  • Data Pipelines
  • ETL
  • Stream Processing
  • MySQL
  • PostgreSQL
  • Amazon Aurora
  • Redis
  • Data Modeling
  • Query Optimization

Testing, Security & Observability

  • JUnit
  • Mockito
  • TDD
  • OAuth 2.0
  • JWT
  • Secure Coding
  • Prometheus
  • Grafana
  • OpenTelemetry
  • Splunk

Engineering Practices

  • System Design
  • Software Architecture
  • Scalability
  • High Availability
  • Fault Tolerance
  • Concurrency
  • Git
  • Maven
  • Agile
  • Code Reviews

05 · Education

Where the foundation came from.

Northwestern University

Degree
Master's in Computer Science
End
Jun 2021

06 · Contact

Let's talk about backend, distributed systems and applied AI.

Email is the quickest route.

Email mepahulsinggh@gmail.com