I build platforms by day, analyse markets by night, and occasionally write code that surprises me. This is my corner of the internet.
I lead platform engineering teams at a global financial institution — but the work I find most alive happens outside the org chart. Building tools that solve real problems, understanding how markets move, and learning fast in public.
The F&O dashboard started as a way to understand options data better. The AI agent project is how I stay sharp on where technology is headed. Neither was planned — both taught me more than any course.
Outside work, I'm a husband, a father to a curious toddler, and someone who thinks the best ideas come from doing, not just reading about doing.
A federated multi-cloud provisioning control plane for Azure and AWS. An app owner answers a 52-question intake; a real 12-rule policy engine gates it — a FAIL genuinely blocks provisioning — then tenancy vending, Terraform, JFrog and Harness run a nine-stage pipeline streamed live over SSE. Every external system sits behind a swappable adapter, so going live is a config change rather than a code change. Deliberately zero LLM calls: policy and config generation are decisions that should be deterministic. Vendors are simulated — no real cloud spend.
Open CloudPro →Multi-tenant Income Tax, GST and TDS filing for chartered accountants — old-vs-new regime side by side, GSTR-2B reconciliation that flags input credit at risk, Form 16 generation, a compliance calendar and a per-client document vault. 421 tests, and CI fails the build on a cross-tenant read, because "we'll remember to filter by tenant" is not a security model. My first isolation harness reported nine breaches — all nine were false positives, caught by adding row-count and random-UUID controls.
Open TaxPro →A stock screener with its own expression language, a backtester that shares the screener's evaluation engine, options-flow analytics over the NSE derivatives bhavcopy, sector rotation and paper-trading books. Screening runs on wide frames — dates × 750 symbols — so a 50-day average is one rolling mean across the whole universe and a full scan lands in ~80ms. User expressions are parsed by a whitelisted AST walker: no eval, and deliberately no LLM in the execution path. Every result is measured against a matched control.
Open TrendRiders →Page operations, native text editing with real paragraph reflow, ink and drawn signatures, OCR to searchable text, cryptographic signing with a visible signature box, and password management via the system keychain. Apple's PDFKit cannot edit page text and has no signature API at all — both were hand-built, by rewriting the PDF content stream and generating PKCS#7 through CMS. The usual workaround is to bundle PDFium; it turned out not to be needed. Swift, no Electron, 81 automated checks.
Read the write-up →A whiteboard built for stylus input: a freeform board, a project/Kanban mode, a kids' drawing studio and eleven brainstorming templates — all a static bundle with no server, no database and nothing to sign into. I rejected Excalidraw (no stylus-only mode, over-strong smoothing) and tldraw (source-available, watermark unless you pay), so the engine is mine. Safari doesn't implement getCoalescedEvents, and ink base width has to be 3–11× the shape stroke width or pressure lands sub-pixel and the pen feels dead. Tested on a real iPad with an Apple Pencil.
Try it — no account →Real-time NSE BhavCopy analysis — OI distribution, PCR trends, institutional money flow and an embedded Claude AI analyst. Built from zero database experience. Shared with a group of trader friends who use it daily.
Open dashboard →An agentic system that queries live options data, reasons about market structure and delivers a morning briefing — a real LLM with tools, not a chatbot. It grew into the TrendRiders platform above, which is where the work continues.
Shipped — now part of TrendRidersCombines Order Blocks, BOS, CHoCH, Liquidity Sweeps and Fair Value Gaps across 4H trend and 15M entry timeframes. Built to trade smarter, not louder — using Smart Money Concepts to identify institutional footprints.
TradingView private scriptMy wife and I like to play Ludo over breakfast. Every app we tried was buried in ads. So I built one — no ads, no accounts, just the game. Sometimes the best reason to build something is the simplest one.
Play →A 14-tool browser-based PDF suite — merge, split, compress, rotate, watermark, protect, convert, reorder pages, chat with PDF, contract analysis and invoice extraction. All core tools run entirely in the browser with no file uploads. AI features powered by Llama 3.3 70B via OpenRouter — free for visitors.
Open PDFCraft →A fully functional enterprise voice AI agent — "Alex" — that handles IT support calls, troubleshoots issues step by step, detects Major Incidents (P1/P2/P3), arranges Field Technician visits, and logs every ticket live to Google Sheets with auto-generated INC numbers and SLA deadlines. Deployed on a real phone number via SIP trunk.
Call +91 94849 56991 →An AI travel companion for when you don't know where to go. Pick a vibe — beaches, mountains, biking, spiritual and more — and Raahi suggests destinations matched to your budget and who you're travelling with (with safety-first picks for solo women), then builds a full day-by-day itinerary, hotels for every budget, and live weather. Self-hosted end-to-end on my own server in Mumbai (Next.js + Postgres + Google sign-in).
Plan a Trip →A patient, voice-first AI study buddy for Indian school children. Kids speak their doubt in Maths, Science or English; Samajh listens via Whisper, thinks it through with Claude, and speaks back through ElevenLabs — using everyday Indian analogies (rotis for fractions, cricket for force) and breaking explanations into small steps. Covers NCERT CBSE Class 5–12 with an optional chapter picker. Behind a Google login + Discord-button admin approval flow to keep API credits sane.
Visit Samajh →Lead India's Platform Engineering and Infrastructure Reliability function, owning end-to-end IT Operations, Site Reliability Engineering (SRE), and large-scale automation programs across Mumbai, Bangalore and Hyderabad. Accountable for on-premises and hosted Datacentre operations, Network, Servers, and Storage, alongside Production Management and End User Computing for thousands of users. Drive Global Platform Engineering initiatives encompassing Azure cloud deployments, DevOps pipelines, Infrastructure as Code (IaC), Messaging platforms, Kafka, SSL, and Webservices. Spearheading the adoption of intelligent automation and exploring how Generative AI and Agentic AI can augment platform operations.
Established and matured the Chennai site, then relocated to Mumbai to build the operation from the ground up — scaling it to a seat capacity of 3,800 and a live user base of approximately 4,500 spanning Front, Middle, and Back Office Operations. Delivered mission-critical technology support for Front Office trading, Sales, and Corporate Banking teams.
Played a pivotal role in one of Barclays' most ambitious India expansion programmes — orchestrating the seamless transition of over 3,500 roles from the UK, US, and Middle East into the newly established Chennai Technology Centre.
Modern infrastructure teams are leveraging Gen AI to automate runbook generation, accelerate incident triage, and synthesize monitoring signals — reducing MTTR and freeing engineers for higher-order work.
Agentic AI systems autonomously execute multi-step IT workflows — from auto-remediation in SRE pipelines to intelligent provisioning of cloud resources — enabling truly self-healing infrastructure at scale.
Combining IaC principles with AI-driven code generation accelerates deployment velocity, reduces configuration drift, and enables policy-as-code enforcement across hybrid cloud environments.
AIOps platforms use ML to correlate alerts, predict failures before they surface, and intelligently route escalations — aligning well with the SRE model and platform engineering disciplines.