AI / ML engineer & builderCOMPASSION-A-THON 3.0Runner-up · IIT Madras

I build the
intelligence.
And the thing
around it.

I’m Dweep Shishodia. I build AI products, machine learning systems, and automations that turn messy problems into working software.

My startups: Paleskies under Matri6 / Brickstone
AI & Automation Intern, GCEC

Meet Paleskies ↘Explore my experience ↗

A FIELD OF POSSIBILITIESMOVE THROUGH THE SIGNAL ↗

01 / My startups

What I’m building.

Two personal startup projects.
Both at the MVP stage.

PALESKIES UNDER MATRI6 MVP IN DEVELOPMENT

Paleskies.

A product image.
A whole new way to show it.

My AI product-video startup under Matri6. I’m developing the MVP around product-image cleanup, guided generation, and final video composition.

Runner-up · Compassion-a-thon 3.0
PARADOX, IIT Madras · 2026

Read the build story ↗

The problem

Creating product videos usually needs a shoot, editing, and multiple tools. Paleskies explores a workflow starting from a single product image.

What I’m building

Co-founded Matri6 and am developing the Paleskies MVP: a pipeline for image cleanup, AI video generation, and final composition.

A decision that matters

Use repeatable templates to guide generation, rather than asking the user to start from a blank prompt. Treat cleanup and final composition as separate steps.

Evidence & status

MVP in development. The Paleskies prototype was runner-up at Compassion-a-thon 3.0 / PARADOX, IIT Madras, 2026.

Visit Paleskies ↗

FROM IMAGE TO MOTION01 — 03

PIPELINE ILLUSTRATION

Clean the product image and isolate the subject, so the product stays at the centre of the creative.

BRICKSTONE / SPATIAL DESIGNCONCEPT SKETCH / MVP FOCUS

MY STARTUP MVP IN DEVELOPMENT

Brickstone.

From imagination
to interaction.

An AI-powered spatial design startup for the built environment. I’m developing the first MVP around interpreting 2D floor plans and creating structured 3D visualisations.

Initial focus: floor-plan intelligence and 3D visualisation.

Read the build story ↗

The problem

It is difficult to experience a proposed space from a 2D drawing, and conventional visualisation takes specialist work.

What I’m building

Developing Brickstone’s first MVP around floor-plan interpretation and 3D visualisation. The product and implementation are still being shaped.

A decision that matters

Start with a focused floor-plan-to-3D workflow as an entry point to a broader spatial design platform. The final technology stack is not yet fixed.

Evidence & status

Personal startup project. MVP in development; the wider interactive-design roadmap is planned work.

02 / Experience

My ventures.
My work with teams.

Building my own startups, learning through an internship, and contributing to AI engineering with other teams.

Co-founder

Developing an AI product-video MVP for D2C brands, from product-image cleanup to video generation and composition.

Explore Paleskies ↗

MVP STAGE

02 / MY STARTUP

Brickstone

Building the MVP

Developing an AI-powered spatial design startup. The initial focus is floor-plan intelligence and structured 3D visualisation.

Explore Brickstone ↗

MVP STAGE

03 / INTERNSHIP

GCEC Global Foundation

AI & Automation Intern

Building AI automations and learning dashboards for the academic team and startup workflows.

See the work ↗

SEP 2026

04 / COLLABORATION

Cordia

Founding AI Engineer

Approached to contribute to Cordia’s AI engineering. Working on the model and memory service and the client discovery chatbot.

See the engineering ↗

2026

03 / Work with teams & clients

Different problems.
Same follow-through.

Internship work, AI collaboration, and client delivery. Explore the connected ML / DL map below for the models and reports.

03 / AI & Automation Internship

GCEC

Useful systems for the day-to-day work.

AI automations and learning dashboards for GCEC Global Foundation’s academic team and startup workflows.

Automation · Dashboards · APIs · Google Sheets

Read the build story ↗

The problem

Academic and startup teams need usable systems around recurring workflows, learning content, and operational data.

My contribution

Building AI automations and learning dashboards as an intern at GCEC Global Foundation, with work spanning academic and startup workflows.

A decision that matters

Build around the team’s existing tools and make the workflow clear enough to use and maintain.

Evidence & status

AI & Automation Intern, GCEC Global Foundation, Jaipur. Internship began in September 2026.

DS / 03CONCEPT DIAGRAM · GCEC

04 / External AI engineering collaboration

Cordia

From a conversation to a clearer workflow.

Approached to contribute to Cordia’s AI engineering, working on the model and memory service and client discovery chatbot.

Gemini API · Supabase · HTTP services

Read the build story ↗

The problem

Teams need to describe how work actually happens before an AI system can help. A generic survey misses the context behind a task.

My contribution

I own the model and memory service: the reasoning backend called by the agent dashboard. I also built the client discovery survey chatbot with Gemini API and Supabase.

A decision that matters

Separate the reasoning and memory service from the dashboard behind an HTTP interface. Keep the discovery conversation focused on the team’s actual tasks.

Evidence & status

Founding AI Engineer at Cordia. My role covers the model and memory service and the client discovery chatbot.

Public role details ↗

DS / 04CONCEPT DIAGRAM · CORDIA

05 / Freelance / client work

SK Traders

A parts business, connected to its catalogue.

A B2B storefront for an e-scooter spare-parts wholesaler, with inventory sync from its accounting workflow.

Next.js · Supabase · n8n

Read the build story ↗

The problem

A spare-parts wholesaler needed a public catalogue connected to the stock information already maintained in its accounting software.

My contribution

Built the storefront and its inventory integration, along with a photo-upload workflow for adding product imagery.

A decision that matters

Keep the accounting workflow as the source of stock information. Show availability indicators on the storefront, while keeping exact counts private.

Evidence & status

Freelance client build. The public storefront is linked below; operational details and client data remain private.

Visit storefront ↗

DS / 05CONCEPT DIAGRAM · SK TRADERS

06 / Research / in development

Mnemra

Memory starts with who can access it.

An agent memory layer exploring typed shards, provenance, and access checks before relevance ranking.

FastAPI · React · OpenRouter

Read the build story ↗

The problem

An agent remembering something does not mean it should be allowed to use it everywhere. Retrieval alone does not settle access boundaries.

My contribution

Co-building a memory architecture with typed shards, provenance, cascade deletion, and a two-lane write gate.

A decision that matters

Apply access checks before relevance ranking. The intended evaluation compares leak rate, format violations, and token cost; results are not yet presented here.

Evidence & status

Research project in development. The architecture and evaluation plan are being developed together.

DS / 06CONCEPT DIAGRAM · MNEMRA

APPLIED ML / DEEP LEARNINGThe ML / DL map.9 projects in forecasting, classification, clustering, and neural networks. Open the map, select a model, and explore its code or report.Python / scikit-learn / PyTorch / BoostingExplore the mapClose the map

01 / MLForecasting

StockSense

Daily sales forecasts for Rossmann stores, with time-based validation and a prediction API.

LightGBM · Polars · Optuna

GitHub ↗Kaggle report ↗Demo ↗

02 / DLMulti-output regression

ThermalIQ

A PyTorch neural network predicting building heating and cooling loads from the Energy Efficiency dataset.

PyTorch · ANN · StandardScaler

Kaggle notebook ↗

03 / MLClassification / ranking

AgroNav

Predicts visit conversion to help prioritise agri-sales territories. Syngenta Hackathon finalist at IIT Madras.

CatBoost · LightGBM · Optuna · SHAP

Project & results ↗Demo ↗

04 / MLRegression

AutoVal

Used-car valuation, from feature engineering and model comparison to a served price estimator. Training and deployment are grouped here.

Random Forest · scikit-learn · FastAPI

App source ↗Training report ↗Kaggle ↗

05 / MLClassification

IBM Attrition

An XGBoost study on IBM’s HR dataset, with explicit evaluation limits and a prediction API.

XGBoost · Feature engineering · FastAPI

GitHub report ↗Kaggle ↗

06 / MLClustering

SmartCart

Customer segmentation with PCA and clustering comparisons. The served model uses KMeans to assign new customers.

KMeans · PCA · Hierarchical clustering

GitHub report ↗Demo ↗

07 / MLClassification

ShopSmart

Purchase-intent prediction from e-commerce session behaviour, using a tuned decision tree and imbalance handling.

Decision Tree · GridSearchCV · SMOTE

GitHub report ↗

08 / MLClassification study

NovaGen

A synthetic-data health-risk classification experiment focused on feature selection, class imbalance, and recall.

XGBoost · SMOTE · GridSearchCV

GitHub report ↗

09 / MLClassification study

CreditWise

An educational loan-approval modelling study comparing algorithms after feature engineering and scaling.

Logistic Regression · Naive Bayes · KNN

GitHub report ↗

More experiments, baselines, and learning notebooksExplore ML-Journey ↗

PROOF OF WORK, IN THE OPEN

Follow the evidence.

Read the code. Open the notebooks.
See what I’m building and sharing.

04 / Recognition & learning

Work that earned
a second look.

Competition results, foundational study, and the learning behind the work.

PRODUCT / 2026

Runner-up

Compassion-a-thon 3.0

PARADOX, IIT Madras

For Paleskies under Matri6Meet the product ↗

APPLIED ML / 2026

Finalist

Syngenta Hackathon

IIT Madras

For AgroNav · Agri-sales territory prioritisation

FOUNDATION / SEP 2026

8.0 CGPA

IIT Madras

BS Data Science & Applications

Foundation Level completed · 24 credits

CONTINUING TO LEARNClaude Code 101 · Anthropic Academy · May 2026Verify certificate ↗

05 / Ways to work together

What can we build?

Bring the problem, even if the brief isn’t clear yet. We can work out what is worth automating, build a useful first version, and take it into production.

AI products & assistants

Build a useful AI feature around your users’ actual workflow: retrieval, model integration, memory, APIs, and the interface around it. Start with one clear use case and define how to check it works.

Workflow automation

Connect the tools your team already uses. Reduce repeated data entry, build operational dashboards, and make the handoff clear enough for people to keep using it.

Applied machine learning

Forecasting, ranking, segmentation, and prediction. Start with the business question, build a baseline, validate carefully, and serve predictions through an API.

06 / Skill set

What I bring
to a build.

I’m Dweep — an AI / ML engineer building Paleskies under Matri6 and Brickstone, both at the MVP stage. I’m also an AI & Automation Intern at GCEC and a BS Data Science & Applications student at IIT Madras.

I work across the model, the data, and the system around it. These are the skills I use, and the projects where you can see them.

Go straight to the evidence ↗

Machine learning & deep learning

Forecasting, classification, segmentation, and neural networks. Choose the model around the problem and evaluation.

Python · scikit-learn · PyTorch · XGBoost · LightGBM · CatBoostExplore the model collection ↗

GenAI & semantic retrieval

Embeddings, semantic search, RAG, memory services, and LLM integration — connected to a usable product.

Sentence-BERT · FAISS · RAG · Gemini · OpenRouterSee Cordia ↗

Data & evaluation

Feature engineering, time-aware splits, imbalance handling, tuning, and explainability. Understand what the score actually means.

SQL · Polars · NumPy · Optuna · SHAPRead the model reports ↗

APIs, automation & deployment

Serve predictions, connect operational tools, build the interface, and hand over a system people can use.

FastAPI · Docker · Supabase · n8n · Google Cloud · RenderSee client delivery ↗

07 / Make the next move

Have a project, a role, or a problem worth working on? Tell me what you have in mind.

Let’s make
something useful.

dweepshishodia57@gmail.com ↗

Start a conversation ↗

I build theintelligence.And the thingaround it.

What I’m building.

Paleskies.

The problem

What I’m building

A decision that matters

Evidence & status

Brickstone.

The problem

What I’m building

A decision that matters

Evidence & status

My ventures.My work with teams.

Co-founder

Building the MVP

AI & Automation Intern

Founding AI Engineer

Different problems.Same follow-through.

GCEC

The problem

My contribution

A decision that matters

Evidence & status

Cordia

The problem

My contribution

A decision that matters

Evidence & status

SK Traders

The problem

My contribution

A decision that matters

Evidence & status

Mnemra

The problem

My contribution

A decision that matters

Evidence & status

StockSense

ThermalIQ

AgroNav

AutoVal

IBM Attrition

SmartCart

ShopSmart

NovaGen

CreditWise

Follow the evidence.

Work that earneda second look.

Compassion-a-thon 3.0

Syngenta Hackathon

IIT Madras

What can we build?

What I bringto a build.

Machine learning & deep learning

GenAI & semantic retrieval

Data & evaluation

APIs, automation & deployment

Let’s makesomething useful.

I build the
intelligence.
And the thing
around it.

My ventures.
My work with teams.

Different problems.
Same follow-through.

Work that earned
a second look.

What I bring
to a build.

Let’s make
something useful.