Harsh Kumar

01

AWS Cost Bot

AWS Cost Intelligence Assistant

A FinOps platform for real-time AWS cost visibility with natural language queries powered by GPT-4o.

Problem

Understanding AWS costs required navigating multiple Console pages, manually correlating data across services, and generating reports by hand every month.

Solution

A chat interface where users ask questions in plain English and get instant, data-backed answers with inline charts — replacing all manual cost analysis workflows.

Budget alerting with hourly threshold evaluation
Automated 9-section PDF reports generated on the 1st of each month
3-tier caching layer: in-memory → Redis → AWS API
ML-based forecasting with confidence intervals and anomaly detection

FastAPIReact 18LangChainGPT-4oAWS Cost Explorerboto3RedisRechartsReportLabVite

02

NoteFlow

Management Approval Notes System

Automates creation, review, and approval of Management Approval Notes from Service Orders across the DishTV group.

Problem

Teams manually extracted data from SO documents, calculated GST in spreadsheets, formatted MA documents by hand, and routed them with no audit trail — causing duplicate entries and calculation errors.

Solution

Upload a PDF/DOC service order, get data auto-extracted via OCR, auto-calculate GST and total exposure, and generate signature-ready PDFs through a structured approval workflow.

Auto-flags high-value approvals exceeding ₹1,00,000
WYSIWYG editor with locked structure but editable content
Full approval chain: submit → admin review → approve/reject
Version control with field-level change tracking

React 18TypeScriptNode.jsExpressPostgreSQLPDFKitpdf-parseJWTMulter

03

Bugzy

Bug Tracker Platform

A bug tracking platform for QA teams working across 12+ device platforms with dynamic platform-specific forms.

Problem

Generic bug trackers didn't capture platform-specific details critical for reproducing bugs — device model, OS version, network type — and had no structured assignment or completion tracking.

Solution

Dynamic form fields that change per platform, a complete issue lifecycle with automatic reopen counting, @mentions, admin-only notes, and direct S3 uploads up to 200MB.

Complete lifecycle: Pending → On Hold → Referred To → Completed
Admin dashboard with 6 summary tiles and weekly trend analysis
Direct S3 file uploads via presigned URLs (images + videos)
Top 10 bug raisers leaderboard with completion rates

React 18FlaskSQLAlchemyMySQLAWS S3AWS RDSJWTRechartsVite

04

Command Center

VZY AI Platform — Autonomous SDLC System

AI-powered SDLC automation that classifies requests via GPT-4 and routes them to 7 specialized agents.

Problem

Managing VZY's SDLC involved manual coordination across teams with handoffs causing delays, miscommunication, and zero pipeline visibility.

Solution

A Master Agent classifies requests by type and priority, pushes them to domain-specific agent queues, and processes them through an 18-state workflow pipeline autonomously.

7 specialized agents: Bug, Feature, UI/UX, Dev, QA, Release, Store
Real-time monitoring dashboard via WebSocket broadcasts
Custom queue server replacing AWS SQS — fully local, zero cloud dependency
Single startup script with PostgreSQL in Docker

FastAPIOpenAI GPT-4ReactPostgreSQLWebSocketDockerAlembicLangChain

05

DishTV.io Smart TV Display

Smart Office Display Management

Centralized AI-powered control of 60+ Smart TVs across 12 departments via natural language commands.

Problem

Managing what each office TV displayed was entirely manual — no centralized system, no priority override, and no way for non-technical staff to control displays.

Solution

A natural language command bar where users type "Show dashboard in sales" or "Emergency alert" — the AI parses intent and pushes content to correct screens via WebSocket in real time.

7 content types with priority hierarchy (Emergency overrides all 60+ screens)
AI engine parses intent and extracts target entities automatically
Real-time content push via WebSocket to individual or grouped displays
Department, floor, and device-level targeting

Next.js 15React 19TypeScriptClaude APITailwind CSSSocket.ioFastAPIPostgreSQLRedis

06

EPG Studio

Electronic Programme Guide Dashboard

Real-time programme guide dashboard for DishTV and D2H — 616 channels, 16 languages, 11 genres.

Problem

No internal tool to browse or verify programme data — teams checked raw JSON files or the on-air signal with no visibility into data freshness or coverage gaps.

Solution

Ingests Gracenote feeds from S3, stores in MongoDB, and serves a searchable schedule browser with Now/Next cards, live progress bars, and health monitoring.

Type-ahead channel search with genre and language filters
Now/Next programme cards with live progress bars (10s refresh)
Health dashboard: server metrics, coverage gaps, missing channel reports
Full-day schedule browsing for any date

Node.jsExpressMongoDB 7AWS S3DockerShell ScriptsRate Limiting

07

Knowledge Base

Knowledge Management Platform

Full-stack knowledge management with article creation, approval workflows, public sharing, and team onboarding.

Problem

Knowledge scattered across emails, chat, and shared drives — no editorial process, no external sharing without full system access, and no onboarding structure.

Solution

Rich text editor with hierarchical categories, two-stage approval workflow, time-limited public share links with analytics, and an onboarding module with read progress tracking.

Two-stage approval: new articles and edits require admin review
Public share links with 7-day expiry and view analytics
Onboarding module with essential articles and progress tracking
Content Partner Forms for structured submissions with status tracking

React 18FlaskSQLAlchemyMySQLAWS S3AWS RDSJWTReact Quill

08

Lighthouse Metrics Suite

Watcho Metrics + VZY Metrics

Two web performance analysis tools providing on-demand Lighthouse audits, historical tracking, and automated monitoring.

Problem

Engineers manually ran Lighthouse in DevTools, screenshotted results, and shared them in chat — no historical baseline, no regression detection, no automated tracking.

Solution

Web interface for analyzing any URL with multi-run averaging, detailed metric breakdowns, historical trend dashboards, and automated cron analysis every 6 hours.

Desktop/mobile modes with multi-run averaging for consistency
Full metric breakdowns: LCP, CLS, TBT, FCP, Speed Index
VZY variant adds IndexedDB warm-up for returning-visitor accuracy
Automated cron analysis every 6 hours with email alerts

FlaskLighthouse CLIGoogle ChromeAWS RDSAWS SESCronExcel ExportChart.js

09

Reco Dashboard

S3 Analytics Dashboard

Self-hosted analytics dashboard syncing CSV data from S3 with auto-generated charts across 5 business data sources.

Problem

Business teams downloaded raw CSVs from S3 and built Excel charts manually every day — no automated visualization, no historical comparison, no shared dashboard.

Solution

Syncs from S3 twice daily, generates aggregated JSON summaries for instant loads, and presents sheet-specific charts tailored to each data source.

5 data sources: CP Activation, Active Base, Deleted Users, Payments, Monthly Base
Date comparison and drag-and-drop section reordering
Kiosk mode for wall-mounted displays with light/dark themes
PDF export for stakeholder reporting

FastAPIpandasChart.jsboto3AWS S3AWS EC2APSchedulerJWTjsPDF

10

Sentry

Release Management System

Release governance and QA tracking for builds across 12+ device platforms under Watcho and VZY.

Problem

Build artifacts shared via chat, test reports were inconsistent spreadsheets, no visibility into platform testing status, and no structured routing for failed builds.

Solution

A 6-stage build pipeline with automatic failed build routing, standardized test report forms, and role-based access with platform-level permissions via Orbit SSO.

6-stage pipeline: Upload → Testing → Events → Reports → Status → History
Standardized test reports with severity/status filters and PDF export
CleverTap event test reports integrated into the pipeline
Platform-level permissions — testers only see assigned platforms

ReactTypeScriptFlaskMySQLAWS S3AWS SESRechartsjsPDFReportLabOrbit SSO

11

Streamix Pro

Industrial Video Converter

Web-based video converter supporting broadcast codecs — ProRes, H.264, H.265, VP9 — with batch processing.

Problem

Content team relied on manual FFmpeg terminal commands requiring technical knowledge of codec flags, with no batch capability and temp files accumulating on disk.

Solution

Drag-and-drop web interface — select format and quality preset, queue multiple files, download from browser. Background FFmpeg subprocesses keep the UI non-blocking.

Supports ProRes, H.264, H.265 (HEVC), VP9, and legacy formats
Batch processing with simultaneous conversion queues
Automated disk cleanup of temporary files
Zero dependencies — no database, no build step, single command startup

FlaskFFmpegPython ThreadingHTML5ProResH.264H.265VP9

12

Timesheet

Timesheet Management System

Timesheet tracking with admin-controlled registration, deadline enforcement, auto-save, and manpower analytics.

Problem

Spreadsheet-based approach had no access control, no validation, no audit trail — anyone could edit anyone's entries, and managers had no consolidated productivity view.

Solution

Admin-approved registration with platform assignment, 2-day edit window with auto-locking, auto-save, and SES email reminders at 24 and 48 hours before deadlines.

Daily task logging across platforms: Watcho, VZY, Other
2-day edit window with automatic locking of older entries
SES email reminders at 24h and 48h before submission deadlines
Monthly manpower overview with resource utilization and Excel export

ReactFlaskSQLAlchemyMySQLAWS S3AWS SESFlask-MigrateJWTGunicorn

13

Tools Portal

Internal Tools Management Dashboard

Centralized hub for all internal tools with health monitoring, role-based access, and usage analytics.

Problem

Tool URLs shared informally via chat — new members didn't know what existed, no way to detect downtime, and no tracking of who had access to what.

Solution

Catalogs all tools with real-time online/offline status via 5-minute health checks, three-tier role system, favorites, and a structured access request/approve workflow.

Automated health checks every 5 minutes with status indicators
Three-tier roles: User, Admin, Super Admin
Access request workflow with optional expiration dates
Admin analytics dashboard with configurable time periods and PDF export

Node.jsExpressMySQLAWS S3AWS SESJWTbcryptVanilla JS

14

VZY Load Test Suite

Distributed Load Testing Platform

Distributed load testing platform that validated VZY's production readiness with 99K concurrent users.

Problem

No infrastructure to validate whether the backend could handle real-world traffic before launch — no test scripts, no user simulation, no standardized reporting.

Solution

12+ k6 test scenarios with a distributed engine that auto-provisions EC2 instances, streams per-worker metrics via SSE, and auto-terminates on completion.

99K concurrent users (11 workers × 9K VUs) with 98.97% success rate
9.11M total requests processed in a single test run
Auto-scales the instance based on VU count
Real-time per-worker metric streaming via Server-Sent Events

k6Node.jsExpressAWS EC2DockerSSECodeMirrorpm2Nginx

Projects Delivered

AWS Cost Bot

NoteFlow

Bugzy

Command Center

DishTV.io Smart TV Display

EPG Studio

Knowledge Base

Lighthouse Metrics Suite

Reco Dashboard

Sentry

Streamix Pro

Timesheet

Tools Portal

VZY Load Test Suite

Tools & Technologies

Frontend

Backend

Database

DevOps & Cloud

AI & Automation

The Journey