Prabinder Singh prabindersinghh

◈ Who I Am

🏭  Founder @ Leorit          The intelligence layer for factories · live in a ₹120 Cr plant · leorit.xyz
🎓  B.Tech CS + Business      Thapar Institute of Engineering & Technology (2024–2028)
🔬  Research                  Physics-Constrained VAE (IEEE SOCC target) · RAG-ST under IEEE review
🏆  Hackathons                2nd Nationally — CBI PSB Hackathon 2026 (₹3,00,000) · 3× Intl Podium
☁️  Ambassador                Google Gemini Student Ambassador
🧠  Philosophy                Execution > Experimentation · Systems > Isolated Models

I don't build models. I build systems around models — systems that solve real execution problems.

Before writing a single line of code for Leorit, I personally executed thousands of manufacturing orders. That operational experience became the data moat. That data moat became the product. Now I'm building AI agents that run what I used to run manually.

◈ Leorit — My Company

Leorit — The Intelligence Layer for Factories

Founder · May 2025 – Present

"Manufacturing, made programmable."

Manufacturing is the last major industry without an intelligence layer. Leorit instruments factories, then makes them intelligent — complete factory systems in live production in 13 days, against the 12–18 months a conventional ERP takes.

13 days

_{empty repository →
live production}

5 weeks

_{complete system,
gate → dispatch}

₹120 crore

_{manufacturer,
operating since 1998}

512 tests

_{on a system the floor
depends on daily}

Live in production right now:

AI reads the supplier invoice at the gate and creates a record for every unit on the truck — human confirmation retained in the loop
Append-only event ledger — nothing can ever be edited or deleted; a correction is a new event; stock cannot go negative
Unit-level QR identity for every sack, drum and carton — 500 raw materials, three departments, trucks arriving with 2,500 sacks at a time
11 role-scoped logins — authorisation enforced at the server, never the interface
Halfway through the build, the owner asked for an entire department that was never in scope — then commissioned a second system

Instrument → Intelligence → Manufacturing Models

01 · Instrument	02 · Intelligence	03 · Manufacturing Models
Every physical object gets an identity. Every action gets a permanent record. The factory becomes fully legible.	Document AI at the gate · plain-language Q&A over the factory's own data · anomaly detection on cost, downtime, material loss	Models trained on real production execution data — not a general model with a manufacturing prompt in front of it
✅ Shipping today	🔄 Shipping now	🧪 In development

Every factory we instrument makes the next one faster and the intelligence sharper. This compounds. Software doesn't.

Stack: TypeScript end-to-end · NestJS · PostgreSQL · React · event-sourced architecture · edge-deployed Moat: 3,000+ orders executed by hand · 150+ operational conversations across five states — before writing a line of code

_{Earlier chapter — Leorit OS + LEO (2025): SaaS manufacturing platform + WhatsApp-native AI ordering agent (Gemini Flash). Piloted with 3–4 manufacturers, one paying customer — sunset; its learnings became today's platform.}

◈ Open Source Contributions

🔦 Headlamp — kubernetes-sigs (CNCF)

Active Contributor · 2026

PR	Status	What It Does
`#5558`	✅ Merged	ReleaseNotesModal — GitHub Flavored Markdown fix
`#5371`	🔄 Closed	Token expiry warning banner + auto-logout (approach discussion)

What the contributions cover:

PR #5558: Fixed GFM rendering in ReleaseNotesModal — merged by maintainers
PR #5371: Token expiry warning system — extended /me endpoint to expose tokenExpiry, built TokenExpiryNotification component with 60s polling + 2-min countdown banner + auto-logout; 4 Vitest tests
Reviewed by @illume (Microsoft · Headlamp lead maintainer)

⚙️ Kubeflow — OTel Integration (CNCF)

GSoC 2026 Project 7: Integrate Kubeflow SDK with OpenTelemetry · 350hrs · Hard

PR	Status	What It Does
`#341`	✅ `/ok-to-test`	Cluster-scoped runtime fallback
`#401`	📬 Under Review	Full OTel layer — `telemetry.py` · 676 tests, 0 skips
`#402`	✅ Merged	Polling validation fix — all backends

Architecture decisions:

Centralized kubeflow/common/telemetry.py
Per-iteration child spans for wait_for_job_status() polling
max_traced_iterations cap to prevent span explosion
opentelemetry-api as core dependency · Verified OTel JSON PoC

Mentors: @kramaranya · @dhanishaphadate · @jaiakash

🔁 Metaflow — Nomad Backend (Netflix)

Nomad Executor Integration · 350hrs

Built metaflow-nomad repo with working backend decorator
Improved nomad_decorator.py + setup.py + example flows
Active PR #3018 in Netflix/metaflow

Mentor: Madhur Tandon

🔓 More Contributions

Project	Contribution
Headlamp (kubernetes-sigs · CNCF)	PR #5558 — ReleaseNotesModal GFM fix (merged). PR #5371 — token expiry warning system. Reviewed by @illume.
MLflow (Linux Foundation AI)	Entity validation + protobuf serialization in Feedback/Assessment; union-type validation, tracing tests, backward compatibility
OpenMined Syft	Persistent dataset discovery bug — sync idempotency analysis; proposed minimal fix
opensec-intelligence	Published npm v3.2.0 — local-first multi-agent security scanner with a prove-fix-re-prove exploit-verification loop

◈ Featured Projects

🥈 BehaviorDNA

Continuous Behavioural Authentication for Digital Banking · 2026

🥈 2nd Nationally — Central Bank of India PSB Hackathon 2026 · ₹3,00,000

Layer	Approach
Behavioural biometrics	BehaveFormer + Siamese networks
Sequence modelling	CNN-LSTM
Fraud scoring	XGBoost (AUC 0.868) + SHAP explainability
Latency	~14 ms ML pipeline vs 150 ms SLA
Trust layer	Hyperledger Fabric IMEI binding

Led a 5-person team. Deployed live. Honest open-set EER reporting over inflated paper numbers.

🔐 OpenSec Intelligence

Local-First Multi-Agent AI Security Scanner · 2026

📦 Live on npm — v3.2.0

Multi-agent pipeline: Scanner → Analyst → Consensus → Fixer
Prove-fix-re-prove exploit-verification loop — findings are demonstrated, patched, then re-attacked
TypeScript · runs entirely local · built on cmdr

🔎 TruthShield Pro

Multi-Modal AI Forensics Engine · 2025

🥉 3rd Place — BIOS Hackathon 2025, IEEE × CSED, Thapar

Layer	Model
Vision	ViT + diffusion artifact fingerprinting
Audio	Wav2Vec2 + ECAPA-TDNN voice clone detection
Text	LLaMA3 forensics pipeline
Engine	Cross-Modal Consistency Scoring
Retrieval	FAISS provenance graphs + explainability

🛡 CyberSentinel

Hybrid Agentic Intrusion Detection System · 2025

🥉 3rd Place — Agentic AI Hackathon, University of Ulster, UK

Hybrid IDS: rule-based + ML anomaly · 92%+ accuracy
MITM · DNS spoofing · ARP attack detection
NLP phishing engine + UEBA drift analysis
Raspberry Pi edge deployment · sub-2s threat blocking

Deployed to real edge hardware. Not just classification — active threat response.

📊 RAG-ST

Retrieval-Augmented Deep Learning · Spatial Transcriptomics · 2025

📄 Paper under review — IEEE ICAIB 2025

9-stage end-to-end pipeline
EfficientNet + FAISS patch retrieval + MLP gene expression prediction
PCA / t-SNE / UMAP interpretability stack

🤖 ProcurenAI

Multi-Agent RFP Automation System · 2025

🏅 Finalist — Vibestate/ByteWars, Derby, UK

Sales · Technical · Pricing · Master autonomous agents
SentenceTransformer SKU matching
4 hours → <5 minutes RFP processing time

🏙 CIVITAS-INTEL

Urban Decision Intelligence Engine · 2026

🥉 3rd Place + Innovation Award — AI for Sustainability Hackathon, Canadian University Dubai

Distributed multi-agent platform for AI-driven city management
Real-time urban decision optimization

🔐 QuantumSight

AI-Augmented Cryptographic Intelligence Platform · 2026

PNB Cybersecurity Hackathon 2026 — Live Jury VC Session

Addresses Harvest Now, Decrypt Later (HNDL) quantum threat
Stack: FastAPI + React 18 + PostgreSQL + Docker
12 pre-seeded demo assets · Full SRS + working prototype

◈ Hackathon Record

Result	Event	Project
🥈 2nd Nationally · ₹3,00,000	Central Bank of India PSB Hackathon 2026	BehaviorDNA
🏅 Top 12 National Finalist	Secure Bharat: AI & Cybersecurity Summit 2026 — ISB Mohali (Punjab Police × CyberPeace Foundation)	QuantumSight
🥉 3rd + Innovation Award	AI for Sustainability 2026 — Canadian University Dubai	CIVITAS-INTEL
🥉 3rd Place	BIOS Hackathon 2025 — IEEE × CSED, Thapar	TruthShield Pro
🥉 3rd Place	Agentic AI Hackathon 2025 — University of Ulster, UK	CyberSentinel
🏅 Finalist	PNB Cybersecurity Hackathon 2026 · Live Jury VC Session	QuantumSight
🏅 Finalist	Indo-Israel International Hackathon 2025	MediFlow Vita
🏅 Finalist	Sabka AI 2026 — TIET-UQ	FairHire360
🏅 Finalist	ByteWars / Vibestate 2025 — Derby, UK	ProcurenAI

9 competitions. 1 national silver (₹3,00,000). 3 international podiums. All in Year 1 to 2 of college.

◈ Tech Stack

🧠 AI / ML

⚙️ Systems & Backend

🌐 Product & Data

◈ Leadership & Recognition

☁️ Google Gemini Student Ambassador

Gemini adoption · Developer workshops Community building

📋 McKinsey Forward Program Fellow

Global cohort Leadership + digital transformation

🏗 CollabSphere Founder

72-member student ecosystem ₹90,000 in projects in week one

🥈 CBI PSB Hackathon 2026 2nd Nationally · ₹3,00,000

BehaviorDNA — behavioural authentication for banking

📍 Innovation Mission Punjab Campus Ambassador

State-level innovation & startup ecosystem

🎙 TEDx · SAR · Virsa Marketing Lead

PR, outreach, data-backed event planning @ Thapar

◈ GitHub Analytics

◈ What I'm Building Toward

Manufacturing AI  →  Leorit: Instrument → Intelligence → Manufacturing Models
                      Live: ₹120 Cr paint factory — 13 days to production · gate → dispatch in 5 weeks
                      Append-only ledger · unit-level QR identity · AI invoice reading at the gate
                      Long-term: models trained on real production execution data — this compounds

Open Source       →  Active contributor: Kubeflow SDK (merged PR) · Headlamp (merged PR #5558)
                      Metaflow Nomad backend · MLflow (Linux Foundation AI) · OpenMined Syft
                      LFX Mentorship 2026 Term 3 — applied with 3 working POCs (otelc gen_ai · KFP pod diagnostics · Headlamp Velero plugin)

Research          →  IEEE ICAIB 2025 paper (RAG-ST) under review
                      Bridging spatial biology + retrieval-augmented deep learning
                      Physics-Constrained Generative VAE for battery degradation — targeting IEEE SOCC (with two professors)

Trajectory        →  From India's garment factories → global manufacturing intelligence platform

◈ Connect

"Most people build models. I build systems around models — systems that change how industries operate."

Navigation Menu