MANISH MAHTO — Operations Engineer & AI/ML Engineer

01 // Technical Projects

ENGINEERED
FROM SCRATCH

DEPLOYED // LIVE

Full-stack behavioral credit scoring engine. Engineered a behavioral credit risk model using alternative data to provide scoring for "thin-file" users — bypassed Vercel's 500MB serverless limit by transpiling an XGBoost model into pure Python via m2cgen, achieving a 94% reduction in deployment size.

View Live → GitHub →

Description

CreditNeverBefore is a behavioral AI scoring engine designed to solve the "Thin-File" problem in the credit industry. Most traditional banks reject young users or small-scale entrepreneurs because they lack a formal credit history (CIBIL score). This project uses alternative data to prove creditworthiness through behavior rather than just debt history. The frontend (Next.js 14, React 18, TypeScript, Tailwind CSS) collects user-friendly inputs and calls a Python (Flask) backend that scores features and returns a 300–850-style score plus decision hints.

Use Case

The primary use case is Micro-Lending for the Unbanked. Target: A college student or a first-time freelancer who needs a small loan for equipment (like a camera or laptop) but has no credit card history.

The Process: Instead of checking a credit bureau, the engine analyzes alternative data points — such as UPI transaction velocity, utility bill consistency, and digital footprint — to generate a "Behavioral Trust Score."

The Outcome: It allows Fintech startups to approve low-risk, high-potential users who are currently ignored by the traditional banking system.

Frontend

Next.js 14 / React 18 / TypeScript

Backend

Flask / Python

Model

XGBoost → m2cgen (Pure Python)

Deploy

Vercel Serverless

Optimization

850MB → ~50MB (94% reduction)

Styling

Tailwind CSS

IN DEVELOPMENT

TRACEVISION

Industrial-grade vision-language model for trajectory-aware spatial understanding. Bridging the gap between global image context and fine-grained human attention patterns for real-time operational monitoring and bottleneck identification in manufacturing.

GitHub →

Description

Engineered to bridge the gap between global image context and fine-grained human attention patterns. Focused on real-time operational monitoring and bottleneck identification in manufacturing environments — applying computer vision directly to the industrial domain.

Use Case

Industrial Quality Control & Workflow Optimization. Deployed in manufacturing settings to track worker movement trajectories, identify production bottlenecks, and provide spatial intelligence that traditional monitoring systems miss.

Core

Vision-Language Model

Focus

Trajectory Analysis

Domain

Industrial Manufacturing

Type

Spatial Intelligence

DEPLOYED

SMART MIRROR AI

AI-integrated mirror built with Python and OpenCV. Real-time APIs for weather and news. Facial recognition system enabling personalized user profiles — ambient, adaptive, intelligent.

Description

A hardware-software fusion project — an AI-powered smart mirror that recognizes faces and adapts its display to individual user profiles. Integrates real-time weather, news, and calendar APIs into an ambient, always-on interface.

Use Case

Personalized Ambient Computing. Each household member gets a tailored dashboard upon facial recognition — weather preferences, news categories, calendar events, and virtual fashion trial, all displayed on a reflective surface.

Language

Python

Vision

OpenCV

Feature

Facial Recognition

APIs

Weather / News

Status

Patent Pending

Type

AI/ML Hardware

RANKED 4TH // HONEYWELL HACKATHON

FIRE EVACUATION AI

Real-time pathing AI using game-engine simulations. Solves complex emergency routing problems under dynamic constraints. Built for the Honeywell National Hackathon — ranked 4th.

Description

A real-time emergency routing system built using game-engine simulations for dynamic exit path optimization. Focused on safest route calculation during high-stress evacuation scenarios with constantly changing obstacle maps.

Use Case

Emergency Evacuation in Large Buildings. When a fire breaks out, traditional exit signs become irrelevant if a corridor is blocked. This AI recalculates optimal escape routes in real-time based on fire spread patterns and crowd density.

Platform

Game Engine Sim

Core

Pathfinding AI

Event

Honeywell Hackathon

Ranking

4th / National

Domain

Emergency Systems

Focus

Real-time Routing

DEPLOYED

B-DOC HUNTER

NLP-powered tool that summarizes dense business documents and extracts sentiment insights for executive decision-making. Cuts through information overload — fast, accurate, actionable.

Description

An NLP document intelligence tool engineered to process dense business documents — contracts, reports, filings — and extract actionable summaries with sentiment scoring for executive decision-making.

Use Case

Executive Decision Support. A CFO receives a 200-page quarterly report. B-DOC Hunter extracts key insights, flags sentiment shifts in language, and delivers a concise brief — reducing hours of reading to minutes.

Core Tech

NLP

Output

Summaries + Sentiment

Target

Executive Decisions

Type

Document Intelligence

02 // Core Systems

WHAT I
BUILD

Systems Management

OPERATIONS
ENGINEER

Leading end-to-end digital transformation of a traditional textile manufacturing unit into a data-driven operation at Amithraj Apparels. Implementing production monitoring systems and operational data capture to identify bottlenecks and optimize resource allocation. Managing vendor relationships, production timelines, and cross-functional teams across factory floor operations.

Process Optimization Data Auditing System Migration B2B Operations Vendor Management

Community Growth

PerFormance Esports — scaling a digital community from zero to:

1K+

Active members in 60 days

Automated tournament management, onboarding systems, and engagement loops built from scratch.

Technical Stack

Python XGBoost m2cgen TensorFlow OpenCV React.js Next.js Node.js TypeScript PostgreSQL MongoDB NLP SQL Flask Tailwind CSS

Cloud & Infra

AWS Azure GCP Vercel Serverless Local LLM

Hardware // Optimized

CPU

AMD Ryzen 7600X

CURVE OPTIMIZED

CURVE OPTIMIZED // 88%

GPU

AMD Radeon 7800 XT

UNDERVOLTED

GPU UNDERVOLT // 94%

Education

B.TECH AI & ML

Jain University, Bangalore

2021 – 2025

7.4

SGPA

Artificial Intelligence & Machine Learning Specialization

03 // Experience Log

WHERE I'VE
BEEN

Aug 2025 — Present

OPERATIONS ENGINEER

Amithraj Apparels — Bangalore, India

Leading end-to-end digital transformation of a traditional textile manufacturing unit
Implementing production monitoring systems and operational data capture
Managing vendor relationships, production timelines, and cross-functional teams
Phasing in high-efficiency technical upgrades without disrupting production

Jan 2023 — April 2023

DIGITAL MARKETING INTERN

Samarya Naturals

Executed SEO-driven campaigns and database management
Designed ambassador programs and college outreach initiatives
Optimized customer acquisition funnels and organic reach

May 2021 — March 2022

FOUNDER & DIRECTOR

PerFormance Esports

0 → 1,000+ active participants within 60 days through targeted outreach
Designed and executed tournament operations including logistics and participant management
Managed end-to-end event delivery: venue coordination, vendor negotiations, timeline management
Automated onboarding systems and engagement loops

04 // Beyond The Screen

OFF THE
CLOCK

Financial Markets

F&O & CRYPTO MARKETS

4+ years analyzing Nifty/BankNifty F&O derivatives and cryptocurrency markets. Managing personal capital through high-volatility cycles using disciplined risk-to-reward ratios, price action analysis, and systematic frameworks. The same pattern recognition applied in engineering, applied to global capital markets.

Price Action Order Flow VIX Tracking R:R Ratios F&O Derivatives DeFi Crypto

Competitive Gaming

ESPORTS

Founded PerFormance Esports. Competitive gaming as a discipline — strategy, reaction time, and team coordination under pressure. Scaled tournaments to 1,000+ participants.

Tournament Ops Community Strategy

Combat Sports

MARTIAL ARTS & BOXING

Discipline through combat. Martial arts and boxing — building mental resilience and physical precision.

Fitness

GYM & HORSE RIDING

Consistent strength training and equestrian pursuit. Physical optimization as a lifestyle — not a routine.

Automotive

CAR CULTURE

Engineering appreciation through automotive design, performance tuning, and the culture of machines.

Tech Hardware

PC OC & ASSEMBLY

Overclocking, undervolting, and building performance-optimized PC systems. Squeezing every drop of efficiency from silicon — Ryzen 7600X curve optimized, RX 7800XT undervolted. Digital architecture born from understanding systems at their core.

Overclocking Undervolting Custom Builds Benchmarking

Research

IT R&D

Deploying local LLMs, setting up autonomous AI agents, engineering serverless architectures. The drive isn't just to build — it's to build with absolute efficiency. Looking at an 850MB dependency tree and seeing 800MB of dead weight.

Local LLMs AI Agents Serverless Optimization