Skip to content
View utsxvrai's full-sized avatar
😎
Learning!
😎
Learning!
  • Primetrace
  • Bengaluru
  • 04:45 (UTC -12:00)

Block or report utsxvrai

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
utsxvrai/README.md

// $ cat ~/whoami.ts
const utsav = {
  role:      "Backend Engineer @ Primetrace — consumer AI apps (Tarot99 & Astro99: 10M+ downloads each)",
  base:      "Bengaluru, India",
  ships:     ["production RAG", "multi-provider LLM orchestration", "payments & subscriptions", "event-driven workers"],
  stack:     ["Node.js", "Fastify", "TypeScript", "Python", "FastAPI", "Redis", "MySQL", "AWS"],
  sidequest: "competitive programming — ICPC Regionalist 2025 (AIR 603)",
  motto:     "if it can fail over, it should fail over",
} as const;

⚡ What I've shipped

🧠 Production RAG platform

Natural-language Q&A over a large community-post corpus.

MySQL ─▶ MinHash dedupe ─▶ BGE-M3 ─▶ Qdrant
        streaming GPU ingestion
  • Re-engineered ingestion to stream on GPU instead of staging huge intermediate files
  • FastAPI + Next.js serving: SSE streaming, source-cited answers, MMR reranking, semantic answer cache
  • Idempotent nightly incremental re-index → prototype to live product

🔀 LLM orchestration layer

AI chat for two apps with 10M+ downloads each.

request ─▶ router ─┬─▶ Gemini
                   ├─▶ Gemma-4-31B   (Bedrock)
                   ├─▶ Qwen3-235B    (Bedrock)
                   └─▶ GPT-5 / Vertex
        provider down? ─▶ automatic failover
  • Chat, classification & summarisation routed across AWS Bedrock, Vertex AI, OpenAI
  • Prompt caching + failover keep chat live through upstream outages
  • Node.js · Fastify · Sequelize · Redis

💳 Payments & subscriptions, end to end

  • Razorpay UPI AutoPay recurring debits
  • PhonePe order reconciliation, Google Play billing
  • Per-country plans
  • Async AWS SQS/SNS workers for renewals, expiry & attribution

📞 Voice AI calling — Salesence (intern)

  • Backend for an AI outbound-calling platform: Twilio + ElevenLabs ConvAI
  • Real-time streaming at ~500 ms with interruption handling
  • Lead pipeline with sentiment detection + WhatsApp automation for 100+ concurrent leads → 90% faster sales response

🌍 Open source

Recently fixing bugs across the TypeScript / Node / LLM-tooling ecosystem.

Repo What
✅ merged prapaa-ai/agav grep_search context lines for an AI coding agent
✅ merged elie222/rakazo blank taught-skill names fall back to the goal
🔍 in review prisma/orm dedupe repeated values in IN lists
🔍 in review sequelize/sequelize nullable TIMESTAMP on MySQL/MariaDB
🔍 in review redis/ioredis malformed RESP no longer leaks into the next reply
🔍 in review bitwarden/clients bw serve send list stale after sync
🔍 in review meteor/meteor failed ChangeStream start no longer crashes the server
🔍 in review langchain-ai/openwiki duplicate front-matter key wiped every field

…plus open PRs in svelte language-tools, yarn berry, isomorphic-git, loopback-next, infisical and more — see all.

🛠️ Things I've built

Project What it is
cacheflow · @cr7bit/cacheflow Dual-layer (in-memory + Redis) Node.js cache with auto-failover. Published on npm.
lfm learn from mistakes — real backend engineering mistakes, one at a time.
GrindTech Prep platform for DSA, system design and backend interviews.
AI interview prep platform Real-time mock interviews: <500 ms speech-to-text feedback, 50+ concurrent sessions, 40% lower AI cost via Redis caching.

🧰 Stack



RAG · Qdrant · FAISS · BGE-M3 · AWS Bedrock · Vertex AI · OpenAI · SSE · WebSockets · SQS/SNS · event-driven systems

🏆 Competitive programming

ICPC
LeetCode Codeforces CodeChef

📈 Activity

> currently: shipping LLM infra in prod, breaking things in staging, fixing them in open source
> open to: interesting backend / AI-infra problems & freelance — DM on X or LinkedIn

Pinned Loading

  1. GrindTech GrindTech Public

    A modern platform designed to help developers master Tech Skills, Interviews, DSA, System Design, and Backend Engineering, powered by AI-driven content and automation. Built for speed, optimized fo…

    JavaScript

  2. Distributed-Cache Distributed-Cache Public

    Simple & powerful Node.js caching library. Dual-layer (local + Redis) with auto-failover. <1ms latency. Published on NPM.

    JavaScript

  3. juet-play juet-play Public

    JavaScript

  4. flight-service flight-service Public

    JavaScript