[00:00.001] infosession start · client=portfolio · ua=$browser
[00:00.018] okauth: rotating refresh token verified
[00:00.042] okconnecting to openrouter · context loaded · 9 projects
[00:00.061] infooperator shivam_rana · senior backend · 3+yrs · surat,in
[00:00.079] okavailability: now · remote · us/eu overlap
[00:00.094]readyscroll ↓ · or run ./ask below to query this portfolio

i build backends
that don't break,
3× faster.

Full-stack developer · 3+ years shipping production-grade AI and SaaS products. I pair production discipline with AI-augmented workflows (Claude Code, Cursor, MCP) to ship features faster than teams 3× my size.

▸ ask my portfolio ↵case study →hire me →
statusavailable · now
specialismNextJS · AI · AWS
timezoneUTC+5:30 · IN-SR
ai stackClaude Code · Cursor · MCP
§ 01ask.my.portfoliolive · powered by open llms via openrouter

don't read it. ask it.

This isn't a chatbot widget. It's a working agent loaded with my CV, every project I've shipped, the stacks I've used, the metrics I've hit, and how I work with clients. Ask anything — pricing, architecture decisions, specific projects, fit for your team. Real answers, no recruiter dance.

~/shivam-portfolio · agent.tsx
open llm · openrouter
agent> response
Hi — I'm Shivam's portfolio agent. I've got his CV, every project he's shipped, and how he works with clients loaded as context. Ask me anything: specific projects, his work, fit for your team, availability, how he uses Claude Code, you name it.
▸
§ 02vitalslast 5 years · cumulative

what numbers look like
when boring code pays off.

[ M-01 ] perceived wait time0% ↓Real-time AI presentation engine — streaming generated slides with sub-3s time-to-interactive.
[ M-02 ] database storage0% ↓Custom HTML normalization for auto-save — stripped redundant Tailwind boilerplate and spatial data.
[ M-03 ] redundant API saves0% ↓Optimized auto-save pipelines to eliminate unnecessary writes while keeping the editing experience seamless.
[ M-04 ] asset throughput0× ↑Parallel document processing with S3 presigned URLs and batch uploads for heavy enterprise decks.
§ 03case.study / presentation-enginedeep-dive · 1 of 9
● AI Saas · In production · Enterprise Ready

Real-Time Presentation & Document Engine

High-performance AI presentation and document generation platform capable of real-time multi-slide streaming, complex multi-sheet Excel ingestion, and pixel-perfect PDF/PNG export. I architected the streaming parser, normalized auto-save pipeline, LLM context compressor, and headless render service.

Next.jsTypeScriptFastAPIExpressPuppeteerMongoDBRedisQdrantS3 / R2Cerebras LLM
time-to-interactive<3s(-85%)
db storage footprint−70%
document ingest speed4x faster
render export latency-60%
auth & payload leakage0 events
[01] problem
Monolithic Payloads, Latency Spikes & Sluggish Canvas UI.

Generating multi-slide presentations led to blank-screen wait times of over 20s. Unsanitized HTML blobs with raw floating-point coordinates (e.g. 142.8934px) bloated DB storage. PDF exports locked node processes and caused layout shifts.

  • →20s+ initial wait for generated slides
  • →Bloated raw HTML payloads crashing DB storage
  • →Node event-loop blocks during PDF exports
  • →Unsanitized HTML leading to token leakage in error paths
[02] architecture
Normalized Auto-Save + Context Compression + Render Pool.

Rebuilt the entire generation and rendering stack around three key pillars: an optimistic HTML normalization engine, a 14-tool agentic backend with Cerebras LLM context compression, and an isolated Puppeteer render pool for pixel-perfect PDF exports.

  • →Normalized auto-save layer(strips bloat & rounds float coords)
  • →14 native LLM tools (HtmlGrep, line-range view, sandbox execution)
  • →Context compression layer via Cerebras / gpt-oss-120b
  • →Dedicated Dockerized Puppeteer render pool (dynamic viewports)
[03] outcome
Sub-3s Slide TTI, 70% Less Storage & 4x Ingestion Speed.

Reduced Time-to-Interactive from 20s+ down to <3s. Database storage per deck shrank 70% by stripping inline class bloat and rounding spatial values. Multi-sheet Excel and multi-asset ingestion became 4x faster with zero export layout shifts.

  • →85% user perceived wait time(TTI under 3 seconds)
  • →-70% DB storage footprint
  • →-50% redundant auto-save calls
  • →4x asset upload throughput
  • →-60% PDF/PNG export render latency
  • →0 auth/token leakage incidents
┌─ arch.diagram / silent-refresh-flow ─┐auth.service.ts · L 1—124
┌─ arch.diagram / deckit-realtime-slide-engine ────────────────────────────────────────── slide_render.ts L 1–180 ┐
                                                                                                                  
  // client                  api gateway / llm              render service             storage / db                
                                                                                                               
                                                                                                                  
  │  React /  │ ──(SSE)──> │  FastAPI Router  │ ──(Job)──> │ Headless Puppeteer │ ──> │ S3 / R2 Bucket │           
  │ Next.js UI│ <──(Diff)─ │  Cerebras / GPT  │ <──(Buffer)│ Pool (Dynamic Dim) │ <── │ Mongo / Qdrant │           
  └───────────┘            └──────────────────┘            └────────────────────┘     └────────────────┘           
        │                           │                                                                             
    optimistic                  context comp.                                                                     
    auto-save                   + line numbers                                                                    
        │                           │                                                                             
        ▼                           ▼                                                                             
   (HTML Norm Layer)      (14 Native Tools SSE)                                                                   
   • round coords          • view / grep html                                                                     
   • strip tailwind        • add/delete slide                                                                     
                                                                                                                  
  // guarantees                                                                                                   
  - <3s time-to-interactive    ✓ verified • streaming markdown parser                                             
  - zero layout shift          ✓ dynamic auto-height & viewport recalculation                                     
  - 70% lower db footprint     ✓ custom inline sanitization & html normalization                                  
  - zero token leakage         ✓ scoped context compression & router model
§ 04trace / how-i-shiplive replay · agentic workflow

how i ship features 3× faster
without breaking prod.

A real replay of how I drive Claude Code through a feature ticket. Spec → plan → implementation → tests → PR — with me in the loop on every architectural decision, not babysitting boilerplate. Press ▸ play below.

▸ claude-code · ticket #105 · html-normalization
0 / 22710%
step 1 / 8
▸ ready.
press play to replay how i shipped
the html-normalization feature with claude code.
§ 05projects / log9 entries · click to expand

production projects, indexed.

#project · descriptionindustrystatus
[#01]
Real-Time Presentation Engine
AI presentation & document platform — real-time SSE slide streaming, normalized auto-save pipeline, LLM context compression, headless render pool.
SAAS / AIin prod[+]
[#02]
Clario
Modern bookmark manager and browser extension for saving, organizing, and discovering websites, notes, images, and more.
PRODUCTIVITY / SAASlive[+]
[#03]
FlowDraw
Collaborative real-time drawing platform for creating and sharing canvases with low-latency interactions and persistent storage.
SAAS / COLLABORATIONlive[+]
[#04]
LoopRoom
Lightweight real-time chat application with temporary rooms that automatically expire when all users leave.
REAL-TIME / SAASlive[+]
[#05]
ReBrain
Personal second-brain platform for saving, tagging, organizing, previewing, and sharing links, tweets, videos, documents, and notes.
PRODUCTIVITY / SAASlive[+]
[#06]
Url-Fetch-LLM
AI-powered semantic search platform for indexing and searching web content from URLs using vector embeddings, intelligent ranking, and relevant snippet extraction.
AI / SEARCHlive[+]
[#07]
Mailmind — AI Email Assistant
AI email assistant that summarizes, searches, and manages emails through natural-language interactions.
AI / PRODUCTIVITYlive[+]
[#08]
Eventsync
Event management system for organizing and managing events with timezone support, update history, and profile-based filtering.
EVENT MANAGEMENT / SAASlive[+]
[#09]
Dacoid
Android fitness application for setting goals, tracking progress, scheduling workouts, and managing fitness activities.
FITNESS / MOBILElive[+]
§ 06stack / matrixordered by frequency · bright = daily

tools i reach for every day.

backend.core// daily
NestJSNode.jsTypeScriptREST APIsGraphQLMicroservices
cloud.infra// weekly
AWS · S3 / EC2 / RDSAWS IVS · LambdaDockerGoogle CloudFirebaseCI/CD
data// daily
PostgreSQLRedisMySQLMongoDB
security// always-on
AIOAuth2 · JWTRefresh Token RotationEncryption-at-rest
frontend// when needed
ReactNext.jsReact NativeCapacitorSCSS · NgRx
ai.augmented// daily
Claude CodeCursorMCP IntegrationsAgentic WorkflowsPrompt Engineering
§ 07history.logcommit graph · 2020 — now

five years of shipping things that didn't break.

Dec 2025 · nowfounding engineer
Deckit AI
// AI · SaaS · remote
  • →Spearheaded real-time AI presentation engine using Next.js 15 and Streaming Markdown — enabled under 3s Time-to-Interactive and cut perceived wait time by 85%.
  • →Engineered custom HTML normalization + sanitization for auto-save — reduced database storage requirements by 70% and redundant API saves by 50%.
  • →Architected a parallel document processing pipeline using S3 presigned URLs and batch uploads — reduced processing time by 60% and increased asset throughput by 4x.
Nov 2025 · nowfull-stack developer
Freelance
// AI · SaaS · client products
  • →Built and deployed production-grade AI applications using Next.js, React, and Node.js with LLMs, vector databases, and real-time streaming.
  • →Designed and scaled backend systems using Redis, PostgreSQL, and MongoDB for efficient data processing, semantic search, and low-latency caching.
  • →Led end-to-end development of AI email, semantic search, and event platforms — owning architecture, performance optimization, and UI/UX.
Oct 2024 · Apr 2025front-end developer
Amruttam Research Institute
// web · product · growth · hybrid
  • →Led front-end development for scalable web applications serving 5,000+ users in the first month.
  • →Implemented conversion-optimized UI/UX improvements that contributed to 2.5x sales growth.
May 2023 · Dec 2023technical analyst
Junomoneta Finsol
// financial research · derivatives
  • →Increased team productivity by 70% through research and new strategies for derivative opportunities.
  • →Analyzed market trends and crafted hedging strategies for options and derivatives.
  • →Led a 10-member technical analysis team focused on options derivatives research.
§ 08hire / shellaccepting projects · now
~/hire/shivam — zsh — 80×24
private
~ $cat availability.json
{
  "status": "available",
  "start": "immediate",
  "engagement": ["contract", "full-time", "fractional-cto"],
  "focus": ["healthcare", "saas", "fintech", "ai-platforms"],
  "timezone": "UTC+5:30" // strong US/EU overlap
}
~ $./hire --tell-me-more
let's ship
something solid.
// the easiest way to reach me — just email.
~ $[click to copy]
~ $phone+91 90236 31602
~ $whereSurat, India · UTC+5:30 · open to remote
~ $