Hi everyone 👋 I’m currently exploring new Senior Backend / Platform Engineering opportunities. If you know of a role that could be a fit, I’d really appreciate an introduction or quick chat (DMs are open). A little about me: 6 years building scalable backend systems across Zenarate, Delhivery, and GoMechanic Designed and led development of authorization systems (RBAC/ReBAC), event-driven architectures, and multi-tenant SaaS infrastructure serving 700k users Strong experience with FastAPI, Kafka, AWS, PostgreSQL, Redis, ClickHouse, and distributed backend systems Architected a real-time stock analytics platform processing live market data for 13,000 symbols with ClickHouse designed for billion-scale time-series analytics Led backend teams across architecture, code reviews, production rollout, CI/CD, and platform scalability initiatives Experienced working across remote teams and async collaboration environments I’m open to: • Remote worldwide opportunities • Europe/Germany onsite roles • Full-time & contract positions Interested in: Backend Platform Engineering • Distributed Systems • Authorization Infrastructure • Event-Driven Systems 🔗 ravisvats.com

Here is an app I built for myself, and I think it can help many of you. I have CKD. Every 3 months, I get my creatinine, urea, eGFR, and potassium checked. Every 3 months, the doctor asks: "What was your last reading?" Every 3 months, I end up flipping through a stack of reports trying to find it. This year I got tired of it. So I built something. ↓ Medical Record is a free web app for anyone managing a chronic condition — or caring for someone who is. Here's what it does: • Upload your lab PDFs or scans (AIIMS-style reports work perfectly) • It reads the collection date from the report and automatically files everything into the right visit • Extracts every test and value automatically — zero manual typing • Charts trends across visits — creatinine over 2 years, HbA1c every quarter, TSH across dose changes • One-screen doctor view: every marker × every visit, exportable as CSV • If a number looks wrong (bad scan, weird spike), you can verify it against the original report or ask the AI why it flagged it Who it's actually for: → CKD patients tracking creatinine, eGFR, and potassium across visits → People with diabetes getting HbA1c every 3 months and wanting to actually see the trend → Thyroid patients whose medication keeps changing based on TSH/T3/T4 → Anyone whose reports are scattered across WhatsApp, email PDFs, and printed sheets in a drawer → The person who walks into OPD with a fat folder of papers and only a few minutes with their doctor It is not a diagnosis tool. It doesn't give medical advice. It simply puts your lab history in order — so you and your doctor can make decisions using the full picture, not just one isolated report. Free to use. Your data stays private to your account. 🔗 ravisvats.com/applications/m… If you know someone managing a chronic condition, share this with them. That's the whole ask.

StudyPlan update — what you can use now • Today This week in one place — flip the mini checklist on your dashboard between today’s pinned tasks and the full week. • Plan your day, not just your week — “Plan today” sits next to “Plan this week,” so you can start with a single day if that’s easier. • AI day plans — brain dump what’s on your mind; AI turns it into 3–5 concrete tasks for that day. • Pick any day — Mon–Sun chips in the planner, with badges showing days that already have tasks. • Manual tasks for today — add a task by hand without using AI. • See what you already planned — opening today’s plan shows existing tasks first, then options to add more. • Smarter dashboard card — progress as “2 / 7 done,” category tags, expand hidden tasks inline, and a quick “Plan today →” action. Try it out: ravisvats.com/applications/s…

Layoffs, job insecurity, and rapidly changing tech make engineers doubt themselves sometimes. But software engineering is still one of the coolest professions. You can sit alone in a room with a laptop and build something that millions of people use. Very few careers offer that kind of leverage and impact. 😎

Given Cursor, do you guys also feel building new applications everyday 😅

Ok, my app got LLM integrated, it is sacred now. 😀 Built an AI weekly checklist for StudyPlan using free Groq Type your messy week in plain English → structured tasks by category Check off on dashboard, track history in analytics For students: less “what should I do today?”, more momentum. Give a visit now. ravisvats.com/applications/s…

Redis learnings for today Redis isn't a database that "uses some RAM" — it IS RAM. Every byte in your key and value is a byte of memory you pay for. No "just add disk." Your data has to fit in memory. So 3 rules: make keys short, strip redundant fields, store nothing twice. Here's the exact change that saved me GBs : 1-second stock candles. Key: stock:AAPL:1s, score = timestamp. Before (~108 bytes): {"s":"AAPL","o":292.48,"h":292.49,"l":292.455,"c":292.465,"v":1373,"dv":null,"vw":292.484,"t":1781276855000} After (~44 bytes): [292.48,292.49,292.455,292.465,1373,292.484] What changed? → s is already in the key → t is already the score → dv was always null → killed the field names: JSON object → packed array Same data. ~60% smaller. At millions of rows/day, that's the difference between a 32GB box and a 16GB one. Your data model IS your infra bill. What's the simplest optimization that saved you the most memory?

Just set up live read replicas on a 700M row ClickHouse prod cluster (28 tables, ~88GB). Here's what actually matters: ClickHouse replication ≠ Postgres. There's no standby mode. You need ReplicatedMergeTree ClickHouse Keeper a second node with a unique replica name. Writes go to primary, parts replicate, replica serves reads. The trap I almost fell into: "snapshot the disk → boot a clone → instant replica." Wrong. Replication tracks parts by name checksum in Keeper, not row values. Cloned parts have different merge history, get marked "unexpected," detached, and re-fetched anyway. Disk image = frozen copy, not a live replica. The only way to reuse disk data: convert primary to Replicated* FIRST, then snapshot primary, then launch replica. Order matters. Conversion is painless — touch convert_to_replicated in each table's data dir, restart. Zero data rewrite. Row counts preserved exactly. Don't forget port 9009 (interserver fetch) — not just 9000. And set <interserver_http_host> or replication silently stalls. Result: 1 primary 2 read replicas, all live-synced. Heavy FINAL queries offloaded. Verify with system.replicas — queue=0, delay=0, active_replicas=2. Solid once you respect how Keeper thinks.

Really love seeing hardware projects, when I am gonna make one 😋

Sometimes I really get confused about what is actually left to prepare for interviews. I mean, the whole day I keep feeding prompts to LLMs and asking them questions. The truth seems to be that what I’m really good at is asking questions to LLMs. 🥲

Been building a stock trading platform lately. A few learnings: • Do deep research before choosing market data providers. Platforms like Polygon and Massive have different plans, rate limits, historical data access rules, and dependencies. Switching later can significantly increase development time. • For storing and querying trade-level data, MySQL alone won't scale well. Columnar databases like ClickHouse are a much better fit for analytical workloads. • Historical market data grows insanely fast. If you're storing 25 years of data across thousands of stocks and ETFs, you'll quickly reach billions of records. Plan hot/cold storage strategies from day one. • Caching is your best friend. Many high-frequency queries shouldn't hit your primary database directly. • Decide early whether your goal is minute-level or second-level data. This single requirement heavily influences architecture, infrastructure costs, and scalability. • WebSockets are essential for real-time market feeds. Polling APIs won't be enough for a responsive trading experience. • Think in microservices when dealing with market data pipelines. The volume and throughput can become too large for a single monolithic application. • Invest in observability early. Tools like Grafana and New Relic help you catch issues before users do. • And finally... weekends are peaceful because the markets are closed 🙂

I’m developing apps with so much energy. It gets ready, and then I get bored by it. Cycle repeats 🥲

What do you guys do when the AWS 1 year free tier ends? New account? PS: Still no paying customers.

So can’t we build middleware’s which just auto suggest which model to use for our queries. @cursor_ai can’t you do that ?

Ok thinking of integrating local llm on my project but Claude suggests it’s better to use Claude Haiku, it will be cheaper. But my learning instincts say go for your own set up 🙈

Ok Study Plan is live today. Try it: ravisvats.com/applications/s… What you can do: Create study goals with targets, timelines, and milestones Run Pomodoro timers (works signed in or as a guest) with a completion alarm See today’s target vs what you’ve done, with a progress ring Track your weekly streak day-by-day (Mon–Sun circles that fill green as you hit targets) View a weekly heatmap of study time Log sessions manually when you forget to start the timer Switch between multiple goals from one dashboard Dive into per-goal analytics and jump back to the right workspace instantly Manage your profile and keep sessions synced when signed in Built for people juggling books, exams, side projects — anything that needs consistent hours, not one-off motivation. Free to use. Sign in to save goals and track progress over time — or try the guest Pomodoro without an account. Would love feedback if you try it 🙏 #StudyPlan #productivity #Pomodoro #buildinpublic #studytips

Start preparing your resume devs 🤠

A monolith is often ill-suited for a system serving multiple use cases. At the same time, maintaining dozens of microservices as a solo developer is equally challenging. What to do ?