Yashwant Chowdary Yarra
Let's talk
Yashwant Chowdary Yarra

Yashwant Chowdary Yarra

Software Engineer Chennai, India

Java & Spring Boot.Built for production.

At TCS, I owned the backend redesign of a Spring Boot service: REST API contracts, downstream integrations, validation and error handling, while keeping existing consumers working. Alongside production work, I build a speech-verification prototype and event-driven RAG pipelines in Python on Google Cloud.

Explore my work Download résumé
Now
Systems Engineer, Tata Consultancy ServicesSince Jul 2024 · 2+ years
In production
Java, Spring Boot, REST APIs, Oracle, AWS S3Plus Angular and TypeScript on the frontend
Applied AI
Python, FastAPI, RAG and LLMsA TCS prototype and Recall, live on Google Cloud
Open to
Backend and applied AI rolesChennai, Bangalore or Hyderabad, or remote

Selected work

A few things I've built, from production services to experiments in applied AI. The thinking behind the code, included.

Production API Platform architectureClient requests cross the highlighted validation boundary, then the controller and service layers in Spring Boot. Oracle access uses tuned SQL. Conditional calls to downstream APIs have timeouts and error mapping. A global exception handler produces a shared error envelope.Spring BootClientGlobal exception handlerShared error envelopeValidationControllerServiceOracleTuned SQLDownstreamAPIsWhen needed:timeout + error mapping

Backend / Production systems

Production · TCS

Production API Platform

REST services backed by Oracle, built and operated in production: validation at the boundary, consistent error handling, tuned queries and resilient downstream integrations.

Read the case study
Large File Upload Platform architectureThe backend creates a multipart upload and signs each part URL. The browser uploads parts directly to S3 in parallel, then sends part ETags and its full-file SHA-256 to the backend. After completion, the backend reads and hashes the whole object and compares digests: a match makes the file ready; a mismatch fails verification.BackendBrowser1. Start upload2. Sign part URLsS33. Parallel PUTs4. CompleteAfter all PUTs+ SHA-256Read full object5. Verify SHA-256Match: readyMismatch: reject

Cloud / Backend engineering

Production · TCS

Large File Upload Platform

Parallel browser-to-S3 uploads using short-lived presigned URLs and retries for failed parts, followed by backend readback to compare the completed object with the client’s SHA-256 digest.

Read the case study
Multilingual Consent Verification architectureWhisper transcribes audio. The transcript and expected statement are transliterated and normalised before alignment and scoring. Accept and reject bands go directly to a verdict; only the uncertain score band takes the dashed LLM path. Both paths produce a verdict and explanation from word-level differences.AudioWhisperExpectedTransliterate+ normaliseAlign+ scoreUncertain score bandLLMAccept /reject bandVerdict+ explainWord diffs

Applied AI / Speech processing

A server-side speech verification service that transcribes, normalises across scripts, aligns spoken text against the expected statement and explains any mismatch in plain language.

Read the case study
Recall: AI Knowledge Assistant architectureFastAPI issues V4 signed URLs for browser uploads to Cloud Storage. Eventarc delivers at least once. A worker claims each pending file atomically before extracting, chunking and embedding it into pgvector and full-text indexes. Success marks the file ready; caught errors mark it failed. Each claim carries a 15-minute lease: while it is live, redeliveries get a 409 so Eventarc retries, and once it expires a retry reclaims a file abandoned by a crashed worker. After three claims the file is marked failed. Queries are embedded with Gemini and combined with query text for hybrid search; retrieved passages feed a Gemini LLM that streams an answer with numbered source citations.BrowserStartV4 URLFastAPISign uploadCloudStorageSigned uploadEventarc: at-least-onceWorker: claim fileChunk + embedReady / failedCrash recovery: 15-min leasepgvector+ full-textQueryGemini embed+ query textHybrid searchVector + full-textPassagesGemini LLMAnswer+ citations

Applied AI / Retrieval

Production · Personal project

Recall: AI Knowledge Assistant

A live knowledge assistant for documents and voice notes: cited Gemini answers, practice quizzes and an agent that asks for approval before creating tasks or saving drafts.

Read the case study

Enterprise work is described with client details kept confidential. Diagrams illustrate the architecture.

Backend depth, applied AI range.

I build software that has to keep working after launch: APIs that validate every input and fail loudly in the right places, services backed by well-shaped data, and cloud integrations that move data safely.

With AI, I keep the model inside deterministic engineering. The speech-verification prototype only asks an LLM about the uncertain band of scores, and Recall answers with numbered references to the passages it used and asks for approval before its agent writes.

Explore my toolkit

Experience

  1. Tata Consultancy Services

    Systems Engineer

    – , Chennai

    I own backend services for enterprise applications in Java, Spring Boot and Oracle, from REST API design and data access to cloud integrations and production support, with Angular work on the frontend.

    • Owned the backend redesign of a Spring Boot service: REST API contracts, downstream integrations, request validation and error handling, while keeping existing consumers working.
    • Improved API response times by cutting database round trips and optimising Oracle SQL queries in Spring Boot services.
    • Built resilient AWS S3 multipart uploads with presigned URLs, parallel chunking, retries for failed parts and SHA-256 integrity checks.
    More responsibilities
    • Built Angular components with TypeScript and Reactive Forms, including input validation and REST API integration.
    • Prototyped a multilingual consent verification service in Python: GPU-accelerated Faster Whisper, text normalisation and word-level alignment to catch missing or altered words, with LLM checks for ambiguous phrases.
    • Performed root-cause analysis of production incidents across services and integrations using AppDynamics and application logs, and shipped fixes through Jenkins CI/CD in an Agile/Scrum team.
    • Reviewed teammates' pull requests for API design, error handling and SQL performance, and documented API contracts and integration guides for consuming teams.
    • Tested services with JUnit and Postman, and upgraded vulnerable dependencies flagged by security scans.
    • Java
    • Spring Boot
    • Microservices
    • Oracle
    • AWS S3
    • Angular
    • TypeScript
    • Python
    • JUnit
    • Jenkins
    • AppDynamics

Education

B.Tech in Computer Science and Engineering (AI & ML)

Vignan's Foundation for Science, Technology & Research

– , CGPA 8.41 / 10

Certifications

Claude Certified Developer

Anthropic, 2026

Google AI Professional Certificate

Google, 2026

B1 Preliminary (PET)

Cambridge Assessment English, 2020

Toolkit

Tools I have shipped real work with, grouped by where they sit in a system. The problem picks the tool.

Backend & APIs

Production at TCS; FastAPI in Recall

  • Java
  • Spring Boot
  • REST APIs
  • Microservices
  • Spring MVC
  • Spring Security
  • Spring Data JPA
  • Hibernate
  • Python
  • FastAPI

Frontend (supporting)

Production at TCS: Angular components and forms

  • Angular
  • TypeScript
  • JavaScript
  • Reactive Forms
  • HTML5
  • CSS3

Data & cloud

Production at TCS: Oracle, S3, EC2; Google Cloud in Recall

  • Oracle
  • PostgreSQL
  • AWS S3
  • Cloud Run
  • pgvector
  • SQL
  • PL/SQL
  • EC2
  • AWS SDK
  • Cloud Storage
  • Eventarc
  • Vertex AI

AI / ML

Prototype at TCS; Recall; published research

  • RAG
  • LLM integration
  • Faster Whisper
  • Vector embeddings
  • Prompt engineering
  • Speech processing
  • Deep learning

Delivery & tools

Production at TCS: JUnit, Postman, AppDynamics and Jenkins deployments; Docker in Recall

  • Docker
  • Jenkins
  • CI/CD
  • Git
  • Maven
  • JUnit
  • Postman
  • AppDynamics

Research and publications

Peer-reviewed work in deep learning for medical imaging, where every model choice has to be justified.

  1. 2024CRC Press (Taylor & Francis)Medical imaging / Transfer learning

    Ensemble of Diverse Pre-trained Convolutional Neural Networks for Enhanced Diabetic Retinopathy Image Classification

    Chapter 1 of Advances in AI for Biomedical Instrumentation, Electronics and Computing (CRC Press, Taylor & Francis)

    View book
  2. 2023AcadloreMedical imaging / Deep learning

    A Deep Convolutional Neural Network Framework for Enhancing Brain Tumor Diagnosis on MRI Scans

    Information Dynamics and Applications, Vol. 2, Issue 1

    View paper

Hiring for backend or applied AI? Let's talk.

Based in Chennai, India. Open to Chennai, Bangalore or Hyderabad, or remote.

Email is the fastest way to reach me. I'm also happy to talk through a system design.