Hello, I'm

Swaraj Reddy

Software Engineer

Backend · Go · PostgreSQL · AWS

I research before I build and measure before I change. On SpotMyJob's job pipeline, that raised throughput 64% with no new infrastructure.

Hyderabad, India · Open to backend, platform, and applied AI roles

Continue to experience

Professional experience.

At DexaMinds, I carry work from service design and data modeling through React UI, AWS deployment, and runtime monitoring under fixed compute and schedule constraints.

Software Engineer

DexaMinds

Hyderabad, India

to Present

  • Started and own the 2 Go microservices behind SpotMyJob's job-data pipeline (source discovery and job processing) that feed its job-search service. Re-sized a transform stage's concurrency on measured database load instead of an assumed CPU count, raising throughput 64% as peak database CPU fell from 99.7% to 65%.
  • Researched before building, from day one: wrote the sourcing research (feasibility, data rights, operating compliance) that set SpotMyJob's job-source path and the decision method both Go services follow; removed 88% of a job-board sweep's requests by proving a list endpoint already returned the data per-item calls fetched, with ETag revalidation answering 83% of board reads with a 304.
  • Shipped React features on the seeker web app (React 18, TypeScript, TanStack Query, Tailwind CSS): Discover filters, keyword search with "Did you mean", onboarding, match-score display, and the location picker, and created 60+ frontend test files.
  • Built two Go services for a client's healthcare platform, document ingestion and clinical extraction on AWS Step Functions (April to June 2026), with Gemini, Vertex AI, Azure OpenAI, and Amazon Bedrock adapters behind a validation gate that flagged uncertain documents for human review; in development, a multi-page lab report went from malware-cleared file to structured result in 62 seconds.
  • Set up Claude Code for both Go repositories, with project rules and hooks that block a commit until its checks pass, and work daily with Claude Code, Codex, and Cursor; AI output is checked against tests, benchmarks, and labelled evaluation sets with measured rater agreement.
Earlier at DexaMinds: Software Engineer Intern, Jun to Sep 2025. Shipped React and TypeScript UI-to-API changes and delivered 15+ frontend and backend production fixes.
Education: B.Tech in Computer Science, GITAM University, 2025. CGPA 8.2/10.

Engineering case studies.

Backend / Healthcare AI

Clinical Document AI

Client Project, Healthcare Document Intelligence

April to June 2026: for a client's healthcare platform, built the document-ingestion service and a clinical document intelligence service, which scanned and validated healthcare files, extracted structured fields with recorded model provenance, and flagged uncertain documents for human review, validated in a development environment.

62s Clean File to Result10.6s p50 Extraction0.2s Service Handoff
  • Built Gemini API, Vertex AI, Azure OpenAI, and Amazon Bedrock adapters behind one provider-neutral extraction gate that checked structure, schema, and a confidence bar before accepting a document.
  • Traced a silent handoff failure between the two services to a queue encryption mismatch, fixed it, and proved the fix with a replay and a fresh canary upload.
How I did it
  • Before the first Go commit, compared LLM providers on the same documents, replaced an OCR-then-LLM design with a single extraction call, and chose the AWS event, workflow, and runtime design against official documentation.
  • Direct-to-S3 uploads, malware-scanned in quarantine and promoted only when clean.
  • Orchestrated extraction as a Step Functions workflow over page chunks: a multi-page lab report went from malware-cleared file to structured result in 62 s with a 0.2 s handoff between services, and the extraction workflow's p50 was 10.6 s over nine development runs.
  • Kept both services independently deployable through versioned events, idempotent requests, an event outbox in the extraction service, SQS dead-letter queues, and stale-job reconcilers.
GoStep FunctionsGeminiVertex AI

AI / Full Stack

ResumeCanvas

AI Resume Builder

Built a full-stack AI resume builder on my own: Next.js 16 and TypeScript on MongoDB, Auth.js with Google OAuth and JWT sessions, 28 API route handlers, and GPT-OSS 120B on Groq for resume review, ATS optimization, keyword analysis, cover letters, and bullet writing.

25 of 50,000 Docs Read1 ms vs 30 ms p5092 to 6 Advisories
  • Indexed the dashboard's resume list on owner and last update: on 50,000 synthetic resumes it reads only the user's 25 at about 1 ms p50 against 30 ms for a full scan, so the list cache that served stale data after edits was removed.
  • Traced build failures that came and went to a Next.js font-loader bug with Google's extensionless font links, reproduced them on demand with mocked font responses, and fixed them with the upgrade that carries the fix.
  • Cut dependency advisories from 92 to 6, none critical, by upgrading Next.js, Auth.js, Mongoose, and transitive packages after reading their release notes, with a test proving the sign-in check fails closed.
  • Found the AI features failing silently after Groq retired their model, moved them to GPT-OSS 120B with errors shown instead of canned answers, and chose JSON mode after strict schemas failed 2 of 24 live calls.
How I did it
  • Every AI route requires sign-in, strips markup and script patterns from its input, and is limited to 50 requests an hour per user.
  • CI runs type-check, lint, 138 unit tests, 142 integration tests against MongoDB, a production build, and 15 Playwright end-to-end tests on every change.
  • Replaced copy-pasted template markup with shared components under snapshot tests recorded first, taking SonarCloud's duplicated new code from 5.07% to 0%.
  • Shipped 5 templates with live preview, Puppeteer PDF export, Zod-validated forms, and public sharing through custom slugs.
Next.js 16TypeScriptMongoDBGroq AI

Tech stack.

Programming Languages

  • Go (Golang)
  • TypeScript
  • Python
  • SQL

Backend

  • Microservices
  • REST APIs
  • Event-Driven Architecture
  • Node.js

Databases

  • PostgreSQL
  • pgvector
  • MongoDB

Cloud & CI/CD

  • AWS
  • Terraform
  • Docker
  • GitHub Actions

AI Engineering

  • LLM Integration
  • Claude Code
  • Codex
  • Cursor

Testing

  • Unit testing
  • Integration testing
  • Contract testing
  • Vitest

Frontend

  • React
  • Next.js
  • TanStack Query
  • Tailwind CSS

Open to software engineering opportunities

Got a role, project, or idea? Let's talk.

Based in Hyderabad, India. Open to backend, full-stack, and AI engineering work across onsite, hybrid, and remote teams, with relocation considered.