AI / ML Engineer

AnjisnuRoy

I turn noise into intelligence — building computer vision, NLP, and LLM systems that run themselves.

Explore my work Download résumé

Descend

/ 01 — The descent begins

I transform data-science prototypes into production-ready ML systems — self-running software across computer vision, NLP and LLMs, backed by a Government of India patent.

Patent Holder

GPA 8.97 / 10

5+ models shipped

99.3% detection

/ 02 — Career timeline

The work so far

scroll to travel

Ellume Technologies

AI Engineer

01 / 03

Jul 2025 — Present
Mumbai, IN

Designed & deployed Ellume Vision — end-to-end solar anomaly detection & forecasting with automated ML pipelines and self-running AI workflows.
Built predictive models for fault detection, degradation, cost & energy forecasting via deep-learning + time-series modelling.
LangChain SLM/LLM reasoning for anomaly summarisation & autonomous ticketing — −30% manual intervention.

Itobuz Technologies

Data Science & AI Intern

02 / 03

Sep 2024 — Mar 2025
Kolkata, IN

Productionised 5+ models incl. LLMs & diffusion — +15–20% accuracy via deep-learning optimisation.
Applied DL to text & image/document data, processed 10,000+ rows of multi-layered training data.
Ran statistical analysis to interpret and validate model-performance results across diverse datasets.

Techno India University

Research Fellow

03 / 03

Apr — Sep 2024
Kolkata, IN

Built a novel XAI / LIME evaluation metric — +20% interpretability, −15% errors.
Solved CV problems with mesh-based image segmentation at 90% accuracy.
Documented model architecture & data pipeline for full reproducibility across research teams.

/ 03 — Selected work

Projects, in orbit

Drag, scroll or use ← → to rotate the carousel.

Production · Featured

Ellume Vision

Solar anomaly detection & forecasting

Self-running AI workflows for large-scale asset monitoring, with LangChain reasoning that summarises faults and opens tickets autonomously.

LangChain Time-series MLOps

crack · 0.993

⬢ Patented

CrackNet

Drone structural defect detection

Raspberry-Pi drone running TensorFlow + OpenCV on the edge — 99.3% crack detection, −50% inspection time.

TensorFlow OpenCV Edge AI

→

Multimodal GenAI

ContentCrafter

AI social-content generation

Multimodal pipeline pairing Mistral 1.3B with Kandinsky Inpaint for real-time image-text synthesis and NLP-driven relevance.

Mistral 1.3B Kandinsky NLP

Research

XAI · LIME Metric

Explainability evaluation

A novel metric for scoring LIME explanations — making model reasoning measurable, +20% interpretability.

XAI LIME Stats

/ 04 — The trophy

Government of
India Patent

Inventor of an AI-enabled modular drone system for automated structural fault inspection — formally recognised by the Government of India.

Design No.

465500-001

Registered

July 2025

Certificate

AI-Powered Crack
Detection Drone

REPUBLIC OF INDIA
THE PATENT OFFICE

/ 05 — Capabilities

The network

Hover a cluster to fire its synapses. Every node is something I build with.

/ 06 — The human

Behind the models is someone who just likes making machines understand the world.

B.Tech in Computer Science & Engineering with a specialisation in AI & Machine Learning from Techno India University, graduating June 2025 with a GPA of 8.97/10.

I care about the unglamorous part — taking a research prototype and making it run itself in production, reliably, on real hardware. From a drone over a cracked bridge to an LLM closing its own tickets.

0.00

GPA / 10

0.0%

Detection acc.

Less manual work

/ 07 — You arrived

Let's build something
intelligent.

github.com/anjisnu07 LinkedIn +91 79080 70818