B.Tech CSE (AI & Data Science), Vasireddy Venkatadri Institute of Technology, Guntur — Class of 2026 · CGPA 7.0/10.

Looking for my first full-time Software Engineer role — backend, full-stack, or AI. Guntur, India · open to relocation · immediately available.

I work on the parts of software nobody notices until they're missing: idempotent webhooks, retries with jitter, race conditions, token budgets.

1. My software runs in production every day. Sole engineer for a CA/CS coaching institute — I design, ship, and operate two systems end to end: PaperCheck AI grades 150+ handwritten answer sheets a day, and my WhatsApp automation platform serves 300+ daily queries.
2. The business case is measured. PaperCheck grades a paper for ₹15–21 against the ₹100–190 human evaluators charge, in minutes instead of 2–3 days.
3. Google merged my code. PR #26884 in google-gemini/gemini-cli, approved by a Google maintainer — picked up from a help-wanted issue.
4. TCS CodeVita Season 13: global rank 2028 of 100,000+ participants — top ~2%.

Sole engineer reporting to the institute's directors. I own two production systems end to end — UI, APIs, data model, deployment, security, operations, and post-release verification.

PaperCheck AI — live

Grades batches of handwritten CA/CS answer sheets in parallel — one streaming AI worker per student — and returns annotated, examiner-style PDFs.

• 150+ answer sheets a day in production; turnaround 2–3 days → minutes, cost ₹100–190 → ₹15–21 per paper
• The LLM only extracts inside schema-constrained JSON; decision logic, a faculty human-review loop, and typed fallbacks keep wrong outputs from reaching users
• Root-caused silent streaming hangs: zero-chunks restart gate + 10-step jittered backoff with transient-vs-permanent error classification — retry when safe, fail fast and loudly when not; zero silent failures since
• Browser-direct streaming keeps multi-minute model runs clear of serverless timeouts; SHA-256-keyed prompt cache (6 h TTL); 50 MB signed-token PDF uploads

WhatsApp messaging platform — source

Event-driven admissions automation on the official WhatsApp Cloud API: FAQs, demo booking, payments, doubt triage to faculty, daily MCQ engine. 300+ daily queries.

• HMAC-SHA256 webhook verification with constant-time compare; fast ACK, async processing; message-id idempotency absorbs Meta's webhook redeliveries
• Full-jitter retries with transient-vs-permanent error classification; WhatsApp's 24-hour messaging rule enforced at a single outbound choke point
• Deterministic-first routing keeps decisions off the model, cutting AI cost ~90%; 16-entity Prisma/PostgreSQL schema; runs credential-free in stub mode

All-in-one platform for CA students: AI tutoring over chat and real-time voice, a notes library, a social feed with DMs and stories, study rooms, leaderboards.

• 70.7K lines of TypeScript — Next.js 16 + React 19, 36 REST API routes, 138 reusable React components
• Security enforced at the database: 34 tables, 85 row-level-security policies, 18 SECURITY DEFINER functions on Supabase Postgres; server-side price derivation eliminates client-side tampering
• Real-time voice mentor on an inline AudioWorklet with ephemeral WebSocket tokens and barge-in; cross-browser sample-rate mismatch solved with dynamic resampling
• Three-model Gemini fallback cascade (a degraded answer beats downtime), SSE streaming chat, per-IP sliding-window rate limits with Retry-After

• Complaint Management System — source: a 15-endpoint REST API in Java 17 / Spring Boot 3.1 with a decoupled React client, built in a three-person team through PRs and code review. Database-level race-condition fix proven by a concurrency test; Hibernate N+1 fix with a regression guard; CI via GitHub Actions and Jenkins (root-caused and pinned a JDK 11 vs 17 runner mismatch).
• Student management portal — source: role-based academic portal with Spring Security and Docker MySQL health checks.

Merged code into google-gemini/gemini-cli — Google's official Gemini CLI.

• PR #26884, merged — fix(core): ignore .pak & .rpa game archives by default, approved by a Google maintainer; stops silent context bloat from game-archive files
• 12 PRs submitted, 3 currently open

Backend
Web
Data
AI
Cloud

If this sounds like your team's kind of work — deevieswar44@gmail.com gets the fastest reply.