I'm Shreyas — a software engineer who builds scalable, dependable systems end-to-end. Today I work in R&D at Meril (Nuvo AI) out of IIT Kanpur Technopark, building the products thousands of people rely on in medical and regulatory work — and bringing AI into them to make those processes faster and more efficient.
I’ve never been good at staying in one lane.
A new stack, a new domain, a problem nobody has handed me yet — that’s where I’m happiest. Healthcare one month, generative AI the next, real-time game servers on the weekend.
What stays the same is the discipline: show up every day, learn it properly, ship it well.
Computer science isn’t just my job. It’s the thing I’d still be doing if nobody paid me — I love it from the bottom of my heart.
Objective“Becoming a High Value Agency and delivering Scalable and Dependable Systems which help Organizations Grow and have Success.”
GPA figures are approximate: there's no single standard for converting Indian grades to a 4.0 scale.
Programming Languages: TypeScript, JavaScript, Golang, C++, Python
Frontend & Frameworks: Next.js, React.js, Node.js, Express.js, Redux, Zustand, Tailwind CSS, CSS, HTML
Backend & Databases: MongoDB, MySQL, Firebase, Prisma, Mongoose, JWT, OAuth Integration, WebSockets, Redis
Tools & Platforms: Git & GitHub, Postman, Webpack, Vite, NPM, Claude Code, AWS
Soft Skills: Active Listener, Collaborative, Supportive, Team Player, Creative Thinker, Communication
Shines Fast work the user needs right now. Simple to build, trace and debug.
Costs Holds a connection for the whole job; slow calls pile up into timeouts.
Shines Slow, bursty or retryable work. A queue soaks up spikes.
Costs Job state, results that arrive later, more moving parts.
Hinges on How long the work takes — and whether someone is waiting on it.
In practice Documents take ~40 s and a thousand people hit Generate at once — so this one went async: a job ID back in ~12 ms.
Shines Handing out jobs: each done once, by one worker. SQS, RabbitMQ, Redis.
Costs Once a message is consumed it’s gone — no replay.
Shines Event streams with many readers, replay and per-key ordering at high throughput.
Costs Partitions, retention and brokers to run and plan for.
Hinges on Is it a task to finish, or an event others will want to read again?
In practice Each job runs once, on one worker, with nothing to replay — a plain queue was plenty.
Shines Rare updates, any client, any proxy. The simplest thing that works.
Costs Wasted requests; latency is only as good as the interval.
Shines Two-way, real-time: chat, cursors, multiplayer.
Costs Stateful connections that make scaling and deploys harder.
Shines One-way server push over plain HTTP, with reconnects built in.
Costs Server-to-client only; connection limits on old HTTP/1.1.
Hinges on Who talks, how often, and through what infrastructure.
In practice Progress only flows from server to user, so it streams over SSE and gets through every proxy.
Shines Transient blips — timeouts, rate limits — with backoff and jitter.
Costs Needs idempotency, and careless retries turn into a storm.
Shines A dependency that’s down, or a user who shouldn’t wait. Circuit breakers.
Costs More errors reach the user; needs a graceful fallback.
Hinges on Is the failure temporary, and is the operation safe to repeat?
In practice Model calls fail transiently, so they retry with backoff — and a job key stops a double-click generating twice.
Shines Stateful things like databases. One bigger box, zero coordination.
Costs A hard ceiling, and a single point of failure.
Shines Stateless services. Redundancy, and elastic capacity on demand.
Costs Load balancing, shared state and consistency to think about.
Hinges on Where the state lives — and how much downtime you can stomach.
In practice Workers are stateless, so they scale out on queue depth and p99 stays flat through the spike.
Shines Relational data, transactions, joins and questions you haven’t thought of yet.
Costs Migrations, and write scaling that takes real work.
Shines Known access patterns at huge scale, flexible documents, key-value speed.
Costs No joins — you model for each query, often with weaker consistency.
Hinges on How well you know your queries today, and how much they’ll change.
In practice Jobs, users and documents are relational — Postgres, with JSONB for the model’s output.
Meril (Nuvo AI Pvt. Ltd.) Kanpur, UP
Laitkor Consultancy Services Pvt. Ltd. & Laitkor Labs Lucknow, UP
Open to new roles, freelance builds and collaborations — especially anything where reliability and scale actually matter.
