Portfolio / 2026

DeepakKumaran T S

AI Engineer, Agentic Systems Developer, GenAI Backend Engineer, Full Stack Developer, Drone Pilot

I build agentic AI systems and the backend infrastructure they run on: multi-agent orchestration, verifiable memory, and GenAI services deployed on Azure and AWS.

View my work

Résumé

yrs production experience

AI engines shipped

open-source platforms

certifications

Selected work

Two open-source agent platforms, and the applied AI work either side of them.

Open source · Multi-agent infrastructure

MassClaw

Infrastructure that lets AI agents find each other and cooperate.

An open-source platform where AI agents discover peers, establish verifiable trust, share memory, and orchestrate multi-agent workflows through management APIs, gateway routing and policy controls.

concurrent agents: 50concurrent agents
modules: 5modules
signed identity: Ed25519signed identity

discovery · identity · trust · memory, federated through one registry

PythonFastAPILangGraphMCPPostgreSQLDockerEd25519

Source

Discovery & registry ranking
Agent registries with automated promotion of reliable agents and demotion of poor performers.
NANDA-native identity & federation
Ed25519 signatures over AgentFacts documents, DID identifiers, and peer-to-peer sync for state reconciliation across distributed networks.
Bayesian trust scoring
Reputation with temporal decay, so stale good behaviour stops counting for ever.
Layered safety controls
Audit trails, prompt-injection detection and policy governance across all five modules.

Python · FastAPI · LangGraph · MCP · PostgreSQL · Docker

Open source · Responsible AI memory

Weavory.ai

Tamper-evident memory for agent swarms in regulated industries.

An MCP-native memory platform letting AI agent swarms maintain verified beliefs, auditable decisions and tamper-resistant histories across banking, insurance and healthcare.

assertions/sec signed & verified: 1,800assertions/sec signed & verified
regulated domains: 3regulated domains
audit trail: Hash-chainedaudit trail

signed · hash-chained · revocable

PythonMCPEd25519FastAPIHash-chained storage

Source

Cryptographically verifiable memory
Signed assertions and hash-chained audit trails make every agent action accountable after the fact.
Revocation by tombstone
Beliefs can be retracted without rewriting history, so the chain stays verifiable.
Fine-grained trust controls
Per-signer and predicate-based validation for sharing knowledge under compliance requirements.
Minimal MCP tool interface
Drops into any MCP-compatible agent framework as a lightweight server.

Python · MCP · FastAPI · Ed25519

AI/ML

AI Supply Chain Demand Forecasting dashboard

AI Supply Chain Demand Forecasting System

An LSTM and Transformer-based forecasting engine predicting demand fluctuations using historical and external market data.

Trained LSTM and Transformer models to forecast SKU-level demand trends.
Designed a Streamlit dashboard for visualisation and real-time data interaction.
Achieved accurate predictions using external signals like weather and holidays.

PythonTensorFlowLSTMStreamlitProphetPandas

Live demo Code

AI/ML

AI subscription churn prediction dashboard

AI Subscription Churn Prediction System

An AI-driven system analysing user engagement and payment patterns to predict churn probability in subscription platforms.

Built an ML pipeline classifying churners from behavioural and transactional data.
Feature engineering and cross-validation reached 92% precision.
Surfaced actionable retention strategies through visualisation dashboards.

PythonScikit-learnXGBoostPandasMatplotlib

Live demo Code

AI/ML

ReadSmart AI – Generative PDF Summarizer & Quiz Assistant

A generative AI tool that summarises uploaded PDFs, generates quizzes, and reads content aloud for faster learning.

Multi-model pipeline for summarisation, question generation and audio narration.
Fine-tuned T5 for contextual summaries and used RoBERTa for answer extraction.
Interactive quiz interface for dynamic comprehension checks.

PythonStreamlitT5BARTgTTSLangChain

Code

Full Stack Web

Sphere – AI-Powered Student Community Platform

An intelligent student ecosystem connecting learners through AI-driven collaboration, real-time chat, events and marketplace features.

Full-stack community platform integrating AI-powered study tools and chat.
Personalised recommendations built with LangChain and OpenAI APIs.
Firebase Authentication and Socket.IO for real-time messaging and event updates.

React.jsNext.jsNode.jsMongoDBFirebaseLangChainAWS S3

Code

View all projects

Recognition

Winner – Hack4Future 2023 (AI & Sustainability Hackathon, Bengaluru)
Top 10 Finalist – National Level AI Innovation Challenge, India
Developed 'AI Supply Chain Forecasting System' using LSTM & Transformer Models
Built 'Sphere' – AI-Powered Student Community Platform with Real-Time Collaboration
Designed & Assembled FPV Drone for Autonomous Flight Path Tracking and Obstacle Detection
Created IoT-Based Smart Tracking System – Nihon Kohden RFID & Cloud Integration Project

References

Deepak has consistently showcased outstanding technical expertise and creative problem-solving ability. His proficiency across full-stack development and AI integration reflects both depth and innovation. He brings a perfect blend of technical precision and design thinking to every project he undertakes.

Marrian MitryProfessor, Career ManagementInstructor & Mentor, Northeastern University · October 30, 2024

It has been a pleasure to see Deepak grow into a versatile engineer with strong analytical and leadership skills. His contributions to academic and professional projects, including AI-powered platforms and cloud automation systems, highlight his ability to merge innovation with practical impact.

Vishal ChawlaProfessor, Web Design & UI/UX EngineeringFaculty Advisor, Northeastern University · November 3, 2024

Let’s connect

I’m open to AI engineering roles from January 2027, and always happy to talk about agentic systems.

Elsewhere

Boston, Massachusetts, USA