Software Engineer  ·  Backend · AI · Cloud Arizona, USA Vol. I  ·  March 2026

The Systems Chronicle

An Independent Record of Engineering & Systems Work

Edited & Published by Sai Arvind Krishnan  ·  Backend Systems · RAG Infrastructure · Cloud Engineering

Available: Open to SWE 2026 roles Location: Arizona, USA AZ --:--:--

About the Engineer

Featured Profile · Arizona, USA

Four Systems Shipped.
Every Metric Moved.

Backend engineer with a bias toward correctness, production observability, and measurable outcomes — across RAG pipelines, transactional Java systems, and cloud-deployed data infrastructure.

≈ 2 min read  ·  4 systems in production  ·  2 active roles

Sai Arvind Krishnan is a backend-focused engineer completing a Master of Science in Software Engineering at Arizona State University, where he serves as Technical Lead for the ASU LX Design programme. He builds systems that hold up under real load.

His work spans an on-premises Java inventory platform that replaced 40% of weekly manual labour, a cloud-deployed RAG system serving multi-tenant clients on Azure, and an enterprise telecom billing audit pipeline processing AT&T and Verizon invoices for TEOCO Corporation.

Across every system he has touched in production, the numbers moved in the right direction: P95 latency down 35%, LLM token costs down 40%, SQL query performance up 25%, auth defects down 30%, and recurring manual overhead cut by 40%. Not roadmap targets — shipped results, with evidence.

0%↓

P95 Latency
Reduction

0%↓

LLM Token
Cost Cut

0%↑

SQL Query
Performance

0%↓

Manual
Effort Saved

Career Dispatches

Current · Dallas TX (Remote) · Aug 2025 – Present

SetuCS.io: How a Single Architectural Change Cut Latency 35% and Halved Token Costs

By Sai Arvind Krishnan  ·  Software Engineering Intern

Identified that sequential Gemini LLM calls were the primary latency bottleneck in the RAG pipeline. Consolidated them into a single batched request with ChromaDB vector retrieval, cutting P95 latency 35% and token costs 40% in Node.js/Express. Re-architected the relational schema to 3NF — FK constraints, composite indexes, query plan tuning — boosting multi-tenant query performance 25%. Deployed the platform on Azure VMs behind a NAT Gateway; integrated Azure Monitor to reduce mean detection time 40%. Secured the API with JWT authentication and Casbin-based RBAC, cutting auth defects 30%. Shipped a React/TypeScript frontend (Vite, Tailwind, TanStack Query) that improved dashboard load times by 30%.

  • Node.js
  • RAG
  • ChromaDB
  • Azure
  • JWT / RBAC
  • React
  • TypeScript
  • PostgreSQL
Capstone · Tempe, AZ · Jan 2026 – Present

TEOCO Audit Pipeline Ingests AT&T & Verizon Invoices with Semantic RAG Evidence Layer

By Sai Arvind Krishnan  ·  Software Engineer

Built Python PDF ingestion pipelines using pdfplumber and PyMuPDF to extract and normalise telecom billing invoices for AT&T and Verizon into auditable PostgreSQL schemas. Implemented cross-document validation workflows — field matching, numeric consistency checks, overcharge detection — emitting fully traceable audit outputs. Added a FAISS-backed RAG layer: chunked extracted text, generated embeddings, and indexed for semantic retrieval, enabling evidence lookup and human-in-the-loop reprocessing via clean Python APIs.

  • Python
  • PostgreSQL
  • FAISS
  • pdfplumber
  • PyMuPDF
  • RAG
Industry · Chennai, India · May – Dec 2023

Inox Tools Replaces Spreadsheets with Java Backend — Recovers 40% of Weekly Manual Labour

By Sai Arvind Krishnan  ·  Software Engineering Intern

Rebuilt Inox Tools Ltd.'s entire vendor and inventory operation from spreadsheets into a Java 17 / Spring Boot backend modelling vendors, SKUs, purchase orders, goods receipts, returns, and inventory ledgers on PostgreSQL with Spring Data JPA. Avoided paid SaaS dependencies; exposed vendor KPIs and procurement state via internal REST APIs. Automated batch jobs (Spring Scheduler) for performance metrics and exception reports, eliminating 40% of weekly reconciliation effort. Introduced ledger-based transaction logic with idempotent updates to permanently eliminate double-counting on PO receipts and returns. Added Bean Validation, centralised exception handling, and SLF4J/Logback structured logging for production traceability.

  • Java 17
  • Spring Boot
  • PostgreSQL
  • Spring Data JPA
  • REST
  • Spring Scheduler
Research · VIT Chennai · Dec 2023 – May 2024

RNN Similarity Model Achieves Semantic Audio Matching That DSP Metrics Cannot

By Sai Arvind Krishnan  ·  Research Intern

Built an end-to-end Music Information Retrieval pipeline in Python using librosa, NumPy, and scikit-learn. Engineered fixed-length audio embeddings via MFCC, Chroma STFT, beat-synchronous segmentation, and statistical pooling. Implemented kNN similarity search with cosine and Euclidean distance metrics. Designed and trained a pairwise RNN similarity model (TensorFlow/Keras) with a 0–1 Similarity Index output, capturing semantic similarity beyond handcrafted DSP metrics. Productionised with PyTest, pickle-based artifact pipelines, and a modular CLI.

  • Python
  • librosa
  • scikit-learn
  • TensorFlow / Keras
  • kNN
  • MFCC

Technical Repertoire

Languages

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

AI & Data

  • RAG
  • LLM APIs
  • FAISS
  • ChromaDB
  • PostgreSQL
  • MongoDB
  • Vector Embeddings

Backend & APIs

  • Spring Boot
  • Node.js
  • Express
  • REST
  • ASP.NET Core
  • GraphQL
  • React

Cloud & Infra

  • Azure (VM · Monitor)
  • AWS (EC2 · S3 · RDS)
  • Docker
  • CI/CD
  • Git
  • Agile

Academic Record

Graduate · Arizona, USA

Master of Science in Software Engineering

Arizona State University — Ira A. Fulton Schools of Engineering

Technical Lead · ASU LX Design

Undergraduate · India

B.Tech. Computer Science & Engineering

Vellore Institute of Technology, Chennai

Hackathon Finalist Nominee

© 2026 Sai Arvind Krishnan  ·  Engineered with semantic HTML, progressive enhancement & genuine craft.