Pankaj KumarDEV_ROLE // PORTFOLIO_V2.0
STATUS: ACTIVE // AVAILABLE FOR WORK
Session Time00:00.00
Section StateHERO
LOC // KOLKATA, INDIA (IIEST SHIBPUR)

PankajKumarBuildingsystemswithcodeandtaste

A software engineer with taste. Specialized in crafting real-time analytics, interactive multiplayer applications, and highly predictive Machine Learning models. Focused on performance, robust architecture, and premium design.

01 // TELEMETRY
LIVE_VERIFIED_DATA
LeetCode Solved
0+
Algorithms & data structures solved
LeetCode Rating
0
Top 23% competitive programming pool
CodeChef Rank
#0
Starters 245 competitive standing
Academic Standing
0.00/ 10
Cumulative CGPA
02 // PORTFOLIO

Featured Projects

SCROLL_VERTICAL
SYSTEM_INIT // drawchain

DrawChain

Multiplayer Real-time Guessing Game

A fast-paced multiplayer drawing and guessing game built completely from scratch. Connect with rooms, draw on a shared canvas broadcasted in real time, and earn points based on speed and accuracy. Features elegant reconnect logic and fair scoring mechanics on player disconnect.

  • Sub-100ms canvas broadcast latency via Socket.IO across concurrent players
  • Reduced client asset payload size by 90% (6.5MB to 675KB), optimizing render speed
  • Session-based authentication and automated leaderboard persistence using NextAuth & Prisma
Next.jsSocket.IORedisPostgreSQLPrismaTailwind CSS
CANVAS_BROADCASTLIVE
Tool: BRUSH // Size: 4px
Player_1:Is it a house?
Player_2:Guessed Correct! (+120pts)
SYSTEM_INIT // ipl-predictor

IPL Win Predictor

Machine Learning Match Situation Model

An interactive ML analytics tool that parses over 1,200 Cricsheet ball-by-ball files to predict live IPL match outcomes. Inputs real-time parameters (wickets, runs, overs, target, current run rate) to yield instantaneous probabilities. Integrates SHAP for modular feature impact explanations.

  • Processed 240,000+ ball deliveries into cricket-specific custom features
  • 75% model accuracy validated on unseen historical test match seasons
  • Deployed explainable web dashboard displaying live probabilities and SHAP features
PythonPandasScikit-learnXGBoostSHAPStreamlit
XGBOOST_PROBABILITY_ENGINEIPL_2026
TEAM_A (MI)35%
TEAM_B (CSK)65%
CRITICAL_FEAT: current_run_rate (+0.14 SHAP)
Overs: 14.2Target: 184
SYSTEM_INIT // civsight

CivSight

Autonomous Civic Infrastructure Management System

A decentralized 5-Stage Agentic Pipeline running on Gemini 3.5 Flash for smart city operations. Automates ingestion, duplication check, community consensus, priority-routing, and side-by-side Gemini Vision verification of municipal repairs.

  • Orchestrated a 5-Agent cooperative flow sharing active ticket state via Firestore
  • Visual dual-image verification detecting fake/duplicate proof-of-work uploads
  • Automated routing to departments (PHED, JVVNL) via maps address resolution
React 19ExpressTypeScriptViteGemini 3.5 FlashGCPFirebase
GEMINI_3.5_AGENTIC_PIPELINEACTIVE
Ingestion Class:POTHOLE (98%)
Severity:8 / 10
SLA:24 Hours (PHED)
Verification: side-by-side vision match complete.
Status: RESOLVED+150 XP
SYSTEM_INIT // yourbooking

YourBooking

Enterprise Conflict-Free Hotel Booking Engine

A robust booking management system that prioritizes precise transactions, avoiding double-bookings during high traffic. Implements transaction locks at the database layer and dynamic availability polling in the React client.

  • Optimistic concurrency control to prevent double-booking conditions
  • Redux Toolkit state orchestration for multi-step reservations
  • Automated check-in / check-out availability transitions
MongoDBExpress.jsReact.jsNode.jsRedux Toolkit
DATABASE_MUTATION_LOCKSCONCURRENCY_TEST
TRANS_890AACQUIRED
TRANS_890BQUEUED_LOCK
HOTEL_ROOM_404OCCUPIED
Engine: MongoDB v6.0CONFLICT_FREE
SYSTEM_INIT // bharatbiz

BharatBiz AI

AI-First Business OS for 63M Indian MSMEs & Kirana Stores

An enterprise-grade POS, GST compliance engine, and business OS built for India's 63 million Kirana stores and MSMEs. Orchestrates 4 autonomous AI agents — Vision (OCR), Voice (Hindi/Hinglish), GST Auditor, and Auto-Reorder — to eliminate manual billing errors, automate compliance, and dispatch WhatsApp invoices natively.

  • 4-agent AI pipeline: OCR receipt scanner, Hindi voice billing, GST auditor & smart reorder
  • Autonomous GST Compliance Auditor auto-corrects HSN mismatches & exports GSTR-1/3B
  • WhatsApp-native invoicing & supplier POs — meeting 70% of Kirana communication where it happens
React 19TypeScriptViteGemini VisionWeb Speech APITailwind CSS
AI_AGENT_PIPELINE4_AGENTS_LIVE
VISION_AGENT (OCR)SCANNING
VOICE_AGENT (हिंदी)LISTENING
GST_AUDITORAUTO-FIX
REORDER_AGENTPO_SENT ✓
GST: CGST+SGST ✓WhatsApp_INVOICE ✓
03 // ARCHITECTURE

System & Pipeline Topology

SVG_FLOW_DRAW_ON_SCROLL
Topology 01 // DrawChain

Real-Time Canvas Synchronizer

Illustrates how player drawing coordinate events are broadcasted, cached, and persisted with sub-50ms latency.

Next.jsClient (UI)Socket.IOServerRedisRoom PubSubPostgreSQLPersistenceHeartbeatCrash Monitor
HTTP/WS Flow Heartbeat Checks
Topology 02 // IPL Win Predictor

Predictive Analytics Engine

Illustrates how raw match logs translate to instantaneous team probabilities integrated with feature explainability.

Cricsheet1200+ CSV FilesFeature EngPandas / NumPyXGBoostModel FitSHAP ValuesExplainabilityStreamlitUI Dashboard
Data pipeline Model & SHAP Flow
04 // EXPERIENCE

Professional Timeline

A chronological summary of engineering roles, client work, and program participations.

B.Tech in Information Technology

IIEST, Shibpur
Aug 2023 – May 2027

Focusing on core computer science foundations, algorithmic analysis, and software engineering methodologies.

  • Current CGPA: 7.77 / 10
  • Key Coursework: Data Structures & Algorithms, OOP, DBMS, Operating Systems, Computer Networks

Machine Learning Intern

Tamizhan Skills (RISE Program)
Jun 2026 – Jul 2026

Developed and evaluated classification and clustering models using Scikit-learn, applying end-to-end ML workflows on real-world datasets.

  • Performed data preprocessing, feature engineering, and model validation
  • Improved predictive performance through systematic experimentation and evaluation
05 // OUTREACH

Let's build something
extraordinary together.

Whether you are screening for placements (Flipkart GRiD, campus drives), evaluating ML capabilities, or looking for high-quality freelance execution, get in touch. Response turnaround typically under 24 hours.