Eshanth Kumar Lal Das — Software Engineer & AI Builder

01 / About

Crafting software that actually matters

I'm a Software Engineer and MSCS candidate at UMass Boston (May 2026) with a deep foundation in full-stack development and a growing obsession with AI systems that solve real problems.

My work spans building a solo national security hackathon platform in 7 days to winning People's Choice at the Paul English Applied AI Institute. I don't just write code — I ship systems.

Currently teaching Operating Systems to 50+ students at UMass while engineering my next product. I care about craft, clarity, and code that stands up to scrutiny.

Production AI Projects

Years at Cognizant

🏆

People's Choice Award

const eshanth = {
  location: "Boston, MA",
  role: "Software Engineer",
  degree: "MSCS @ UMass Boston",
  available: "July 16, 2026",
  workAuth: "F-1 OPT · 3 years",
  superpower: "Ship AI fast",
  currentlyDoing: [
    "Teaching OS @ UMass",
    "Building AI products",
    "Grinding LeetCode",
  ],
  openTo: "SWE / AI Engineering roles",
};
 
// Let's build something great
eshanth.hire();

02 / Skills

Tools of the trade

⟨/⟩ Languages

JavaPythonTypeScriptJavaScriptSQLCC++

⚙ Frameworks

ReactSpring BootFlaskFastAPIDjangoNode.jsHibernateJUnit

◈ AI & LLMs

Groq APILlama 3Gemini APINLPPrompt EngineeringMediaPipeOpenCVExplainable AI

⬡ Databases

PostgreSQLMongoDBFirebaseFirestoreNoSQL

☁ Cloud & Tools

AWS EC2/S3GitMavenAgile/ScrumLinuxStreamlit

◻ Concepts

OOPData StructuresAlgorithmsREST API DesignComputer VisionOS

Technical Proficiency

Python92%

Java88%

React / TypeScript85%

Flask / FastAPI87%

AI / LLM Integration90%

Computer Vision82%

Spring Boot78%

SQL / PostgreSQL84%

Skill Radar

03 / Experience

Where I've built things

Teaching Assistant

University of Massachusetts Boston · Boston, MA

Jan 2026 – Present

Support 50+ students through Operating Systems & Digital Design — covering process synchronization, memory management, concurrency, and scheduling in C and C++.
Conduct code reviews and debugging sessions, developing the ability to read unfamiliar code quickly and communicate solutions clearly across different technical backgrounds.

Software Engineer — GenC Program

Cognizant · Hyderabad, India

Jun 2022 – Jun 2024

Completed enterprise engineering program covering Java, Spring Boot, REST APIs, SQL, and Agile methodologies through structured coursework and project-based learning.
Built full-stack application components using Java, Spring Boot, and React — applying REST API design, relational database concepts, and OOP in a professional engineering environment.
Worked in Agile/Scrum workflows alongside engineers at a global tech organization, developing professional habits around testing, documentation, version control, and collaborative development.

Machine Learning Intern

Brainovision Solutions India Pvt. Ltd · Hyderabad, India

Jul 2020 – Aug 2020

Built ML models and data pipelines from scratch — preprocessed real-world datasets, trained and evaluated models achieving 90%+ accuracy, and accelerated data processing speed by 35%.
Delivered a working ML prototype for client presentations — first experience shipping something that real stakeholders evaluated.

04 / Projects

Things I've shipped — hover to flip

◈ SCSP National Security Hackathon 2026 · Solo

SENTINEL — AI Threat Intelligence

Real-time conflict prediction fusing 5 live intelligence signals into a 0-100 threat score. Llama 3 70B generates CIA-style briefs in <5 seconds. Validated against Ukraine 2022 — 72hr early warning.

PythonFlaskReactGroq APILlama 3Leaflet.js

Hover to see how I built it

How I built it

Architecture: 3-layer system — APScheduler pulls data every 30 mins, caches to JSON; Flask REST API serves it; React 18 + Leaflet.js displays it.

Data sources: GDELT (65K news sources), OpenSky (live aircraft), USGS (seismic), Yahoo Finance (defense stocks), ACLED (conflict events).

AI Layer: Groq + Llama 3 generates threat briefs, powers a live chatbot, and runs a claim verification engine with credibility scoring.

Built solo in 7 days.

View on GitHub

🏆 People's Choice — Paul English AI Institute 2025

AI Diabetes Distress Detection

Analyzes patient transcripts to detect hidden emotional distress. Classifies Low/Medium/High, 0-100 score, evidence span extraction. Hybrid LLM + heuristic, validated against DDS-17 with nursing PhD researchers.

PythonGemini APIReactTypeScriptNLPXAI

Hover to see the impact

The impact

Problem: Patients managing chronic disease often hide distress. Clinicians can't catch what isn't said.

Solution: Gemini API analyzes transcript language and surfaces latent emotional cues with evidence spans showing exactly which phrases triggered classification.

Why hybrid? The heuristic layer validates LLM scores — every decision stays explainable for clinical review.

Responsible AI: Supportive recommendations only, no clinical conclusions. PDF export for physician handoffs.

View on GitHub

◈ Full-Stack AI Application

Personalized AI Recommender (XAI)

Full-stack movie recommender where every recommendation comes with a natural language explanation. Multi-step survey, Firebase auth, smart caching for offline support.

PythonFastAPIReactFirebaseGroq APITypeScript

Hover for technical deep-dive

Technical deep-dive

Core idea: Every recommendation has a paragraph explaining why it fits your taste — built from your survey responses.

Stack: FastAPI → Groq + OMDb API for metadata → Firestore for user data → React/Vite + Firebase Auth.

Smart caching: localStorage minimizes Firestore reads and keeps the app functional offline. Results are shareable via social platforms.

Key challenge: Prompt engineering to get consistent, specific explanations that reference the user's actual ratings.

View on GitHub

◈ Computer Vision System

AI Exam Proctoring System

Real-time exam monitoring with MediaPipe FaceMesh + OpenCV. Detects head turns, gaze deviation, face absence, multiple people. Live Fair/Cheating status with color-coded bounding box.

PythonOpenCVMediaPipeNumPyStreamlit

Hover for detection details

How detection works

Head turn: Nose landmark x-position relative to face bounds — deviations beyond threshold trigger alert.

Eye gaze: Iris landmark positions detect left/right deviation. Looking down (writing) is intentionally allowed.

Multi-person: Face count per frame — more than one face = instant flag.

Live UI: Green bounding box for fair, red for detected cheating. Runs with one command: streamlit run proctoring_system.py

View on GitHub

Eshanth Kumar
Lal Das

Crafting software that actually matters

Tools of the trade

Where I've built things

Things I've shipped — hover to flip

Contribution graph

Academic foundation

Always leveling up live

Let's build something great

Eshanth KumarLal Das

Crafting software that actually matters

Tools of the trade

Where I've built things

Things I've shipped — hover to flip

Contribution graph

Academic foundation

Always leveling up live

Let's build something great

Eshanth Kumar
Lal Das