Open to internships & SWE roles•Bhubaneswar, India (Remote OK)

12,000+ users in prod•650+ LeetCode•9.54/10 CGPA

I build full-stack and AI systems
used by 12,000+ people.

Mrinall Samal

Full-Stack Engineer · AI/ML and Linux Systems

B.Tech CSE @ SOA/ITER (CGPA 9.54/10) · prev. Wipro AI/ML

iters.live maintainerSIH 2026 Top 50GCP Diamond Rank

|engineering-profile.md

Bhubaneswar, IN · Remote OK

I engineer full-stack systems and production AI agents that people actually use. From high-traffic campus infrastructure to in-browser sandboxes and offline emergency tools:

iters.live — Campus hub for 12,000+ students. 100K+ daily API hits, Redis cut DB load ~70%, 99.9% uptime. Shipped and maintained solo.

CodeMentor AI — Agentic coding tutor on a 4-tier model cascade. Monaco + Pyodide in-browser runner, 17 synced visual steppers, ~600 green tests.

AapdaSetu — Zero-auth disaster SOS with routing that avoids live Sentinel-1 satellite flood polygons. SIH 2026 Top 50 nationally.

Sanket — Anti-scam shield over WhatsApp & web. 12 offline signatures + 4-tier LLM failover, plain-language vernacular verdicts, 1930 escalation.

GCP DiamondTop 1% globally

650+ LeetCode314E · 250M · 60H

IIT BombayRank 6

CGPA 9.54SOA/ITER

statusopen for SWE & AI roles

B.Tech CSE '28 · prev. Wipro AI/ML

See my projects Resume PDF

|guest@mrinall: ~ — zsh

guest@mrinall:~$

zshguest@mrinall:~

helplscat bio.txtsnake

12:00:00

TRY CLI:

Bhubaneswar, Odisha (remote OK). Open to full-stack, AI/ML, or backend roles.

ResumeGitHub LinkedIn Email

$ cat about.md

~/about

Mrinall Samal

Full-Stack · AI/ML · Linux Systems Developer

I build production systems used by real people. iters.live became my campus's primary platform within a semester, processing ~100K API requests/day; CodeMentor AI delivers an interactive developer workspace with 17 synchronized execution steppers and compiler-judged challenges; and Amma Odisha enables Bhubaneswar citizens to report civic issues mapped by ward with full public accountability. None of these are throwaway demo scripts; they are maintained and monitored.

Day to day I engineer in React and Next.js on the frontend, Node.js, PostgreSQL, and Redis on the backend, with TensorFlow and PyTorch for ML pipelines, and enough Rust and ACPI/WMI to build Linux system tools.

Interned at Wipro as a Software Engineer (AI/ML) during my early coursework at SOA, optimizing production AI/ML pipelines for Fortune 500 clients (Mar–Sep 2025).

educationB.Tech CSE · Siksha 'O' Anusandhan Univ. (2024–2028)

cgpa9.54 / 10.0

locationBhubaneswar, Odisha, India (Remote OK)

focusFull-Stack · AI/ML · Backend · Linux Systems

Get in Touch →

Engineering Edge & Principles

01 /Real production impact

The numbers in the hero aren't marketing copy. They're systems I operate. When something breaks at 2am, I'm the one who fixes it.

02 /Full-stack breadth

Frontend, backend, ML pipelines, cloud infra, distributed systems, Linux system programming. I contribute across the stack without needing hand-holding.

03 /Proven competitive record

Selected Top 50 nationally in Smart India Hackathon (SIH) 2026, placed 6th nationally in IIT Bombay's Geo AI Challenge, reached finalist spots at EY case competition and MumbaiHacks, and hold Diamond rank on Google Cloud Skills Boost (top 1%).

04 /Open source & systems core

Linux ASUS Armoury Crate tool and an AI Framework in Rust, built for real Linux users and developers globally. Led university coding club.

open-to: Full-Stack · AI/ML · Backend · SWE · Remote / Bhubaneswar

// experience.log

Software Engineer Intern — AI/ML

Wipro

Mar 2025 – Sep 2025 · Remote
(part-time, alongside early coursework)

▸Built TensorFlow and PyTorch pipelines for Fortune 500 clients: data ingestion, feature engineering, distributed training, REST serving.
▸Optimized deployed models with INT8 quantization and batched inference for production latency budgets.

## selected wins

5 of 9 records

DateMilestone / OrganizationScope & Context

2026Smart India Hackathon (SIH) 2026 — Top 50Selected among the top 50 teams nationally in SIH 2026

Mar 2025 – Sep 2025 · RemoteSoftware Engineer Intern — AI/ML, WiproProduction AI/ML pipelines for Fortune 500 clients

Dec 2025EY Business Case Competition FinalistFinalist among 500+ elite teams

Dec 2024 · IIT BombayNational Geo AI Competition — Rank 66th nationally in AI/geospatial ML challenge

Nov 2025MumbaiHacks 2025 FinalistIndia's biggest student hackathon

// tech-stack.json & inventory

tech-stack.json

{

"languages": ["Java", "Python", "Rust", "TypeScript", "JavaScript", "SQL", "Bash", "C++"],

"frontend": ["Next.js", "React", "Tailwind CSS", "Framer Motion"],

"backend": ["Node.js", "Express.js", "REST APIs", "WebSocket", "JWT/RBAC"],

"databases": ["PostgreSQL", "Redis", "MongoDB"],

"ai_ml": ["TensorFlow", "PyTorch", "Scikit-learn", "OpenCV", "Hugging Face", "NLP"],

"systems": ["Linux", "ACPI/WMI", "DBus", "Docker", "Nginx"],

"cloud": ["Google Cloud Platform", "AWS", "Vercel", "GitHub Actions"]

}

compact inline view

used-in.map

// where each skill shows up in production

TypeScriptSanket Scam Shield · Amma Odisha · mrinall.dev

Next.jsSanket Scam Shield · JanSeva AI · CodeMentor AI · mrinall.dev

ReactAmma Odisha · BhoomiSahayak · AapdaSetu · Bhūmi Ledger

Node.jsiters.live · Gig Income Proof · BhoomiSahayak · AapdaSetu

Express.jsiters.live · Gig Income Proof · AapdaSetu · SevaSaathi v2

PythonLexa Legal AI · GenAI DLP · AapdaSetu AI engine · Wipro AI/ML

PrismaAapdaSetu · JanSeva AI

SupabaseAmma Odisha · Gig Income Proof · CodeMentor AI

Firebaseiters.live

Redisiters.live · CampusArena

PlaywrightBhoomiSahayak · iters.live portal sync

SolidityVGDP ZK Proof · Bhūmi Ledger

RustAI Framework (Rust) · ASUS Linux Tool

ACPI/WMIASUS Armoury Crate for Linux

MapLibre GLAmma Odisha

LeafletAapdaSetu GIS routing

TensorFlowWipro AI/ML pipelines

Dockeriters.live · Lexa Legal AI · GenAI DLP

// categorized skill index

LANGUAGESJava · Python · Rust · TypeScript · JavaScript · SQL · Bash · C++

FRONTENDNext.js · React · Tailwind CSS · Framer Motion

BACKENDNode.js · Express.js · REST APIs · WebSocket · JWT/RBAC

DATABASESPostgreSQL · Redis · MongoDB

AI / MLTensorFlow · PyTorch · Scikit-learn · OpenCV · Hugging Face · NLP

SYSTEMSLinux · ACPI/WMI · DBus · Docker · Nginx

CLOUDGoogle Cloud Platform · AWS · Vercel · GitHub Actions

$ git log --proof-of-work

Proof of Work

I run most of these in prod. Short version up front — open build notes on any project for why I built it and what broke.

7 deployments · featured · +7 more · +4 minor

origin/main/deployment tree(7 releases)

[01]/codementor-ai

OPEN SOURCE

CodeMentor AI — Agentic Coding Tutor

Chat with a tutor, run its code right there. 17 visual steppers + spaced repetition.

Live Deployment ↗Case study →

// why: Chat, runner, and visualizer lived in 3 tabs — nothing showed what actually executes.

Chat + run in one place: every tutor answer has a runnable Monaco editor attached
17 synced sims: BST to system design, line-highlighted with the tracer
Never dead: 4-tier model cascade ends in offline mode, ~600 tests green

Next.js 14TypeScriptSupabaseMonacoPyodideTailwind

// why this project?

Learning to code usually means three tabs: a chat window, somewhere to run code, and an algorithm visualizer. Nothing connects the teaching conversation to what actually executes on screen.

// what does it do?

An agentic coding tutor on a Groq-first cascade (Groq → Gemini → OpenRouter → offline). Every answer ships with a runnable Monaco editor — Pyodide in a Web Worker locally or Wandbox server-side. 17 steppers stay synced to the tracer, tied together by a 59-topic curriculum DAG.

note: Hardest part was keeping 17 sims in sync with the tracer without jank; moved docs out of a 2.88 MB bundle after it hurt load.

// results

Simulations from BST up to system design stay in sync with the code tracer's line highlighting. SM-2 spaced repetition schedules reviews from how you actually performed, not just what you clicked. The complexity profiler curve-fits measured runtimes against the target O-class instead of trusting the model's claim. Docs moved out of a 2.88 MB client bundle into server-rendered JSON entries, backed by six RLS-enabled Supabase migrations. About 600 tests in 81 suites are green.

August 2026

branch:release/codementor-ai

Sole developer · AI/ML

verified telemetry

●81 test suites

●17 algorithm simulations

●188 doc-set reference library

links & access

code-mentor-v2.vercel.app

[01]/codementor-ai

OPEN SOURCE

CodeMentor AI — Agentic Coding Tutor

Chat with a tutor, run its code right there. 17 visual steppers + spaced repetition.

Live Deployment ↗Case study →

// why: Chat, runner, and visualizer lived in 3 tabs — nothing showed what actually executes.

Chat + run in one place: every tutor answer has a runnable Monaco editor attached
17 synced sims: BST to system design, line-highlighted with the tracer
Never dead: 4-tier model cascade ends in offline mode, ~600 tests green

Next.js 14TypeScriptSupabaseMonacoPyodideTailwind

// why this project?

Learning to code usually means three tabs: a chat window, somewhere to run code, and an algorithm visualizer. Nothing connects the teaching conversation to what actually executes on screen.

// what does it do?

note: Hardest part was keeping 17 sims in sync with the tracer without jank; moved docs out of a 2.88 MB bundle after it hurt load.

// results

Simulations from BST up to system design stay in sync with the code tracer's line highlighting. SM-2 spaced repetition schedules reviews from how you actually performed, not just what you clicked. The complexity profiler curve-fits measured runtimes against the target O-class instead of trusting the model's claim. Docs moved out of a 2.88 MB client bundle into server-rendered JSON entries, backed by six RLS-enabled Supabase migrations. About 600 tests in 81 suites are green.

January 2026

branch:release/amma-odisha

Full-stack developer · Civic Tech

verified telemetry

All Bhubaneswar wards mapped●

Live at ammaodisha.me●

Admin moderation workflow●

links & access

ammaodisha.me github.com/MrinallSamal-byte/Ammaodisha

[02]/amma-odisha

LIVE

Amma Odisha — Civic Issue Reporting Platform

Ward mapping, public issue reporting, and local governance transparency.

Live Deployment ↗GitHub ↗Case study →

// why: No easy channel to report a civic issue or know who is accountable.

Report by ward: file against your actual ward, see who represents you
Public resolution: status visible so follow-up needs no contacts
Real GIS: ward boundaries from GeoJSON, rendered with MapLibre GL

ReactViteTypeScriptSupabaseMapLibre GLTailwind CSS

// why this project?

Citizens had no easy channel to report civic issues or understand local ward-level governance and representative data.

// what does it do?

Civic-tech platform for Bhubaneswar citizens to report public issues, see ward-representative data, and track resolution. Admin triage, ward mapping with Maps API, public dashboard. Open source.

note: Ward-boundary data was messier than the code — GeoJSON cleanup took longer than the map itself.

// results

Citizens can file an issue against their actual ward and see who represents them. Resolution status is public, so follow-up no longer depends on knowing the right office. Admins triage through a dedicated moderation workflow. Under the hood it's a Vite + React SPA on Supabase (Postgres with row-level-security policies and Edge Functions), deployed on Vercel, with ward-boundary data loaded from a GeoJSON dataset and rendered with MapLibre GL.

[02]/amma-odisha

LIVE

Amma Odisha — Civic Issue Reporting Platform

Ward mapping, public issue reporting, and local governance transparency.

Live Deployment ↗GitHub ↗Case study →

// why: No easy channel to report a civic issue or know who is accountable.

Report by ward: file against your actual ward, see who represents you
Public resolution: status visible so follow-up needs no contacts
Real GIS: ward boundaries from GeoJSON, rendered with MapLibre GL

ReactViteTypeScriptSupabaseMapLibre GLTailwind CSS

// why this project?

Citizens had no easy channel to report civic issues or understand local ward-level governance and representative data.

// what does it do?

Civic-tech platform for Bhubaneswar citizens to report public issues, see ward-representative data, and track resolution. Admin triage, ward mapping with Maps API, public dashboard. Open source.

note: Ward-boundary data was messier than the code — GeoJSON cleanup took longer than the map itself.

// results

Citizens can file an issue against their actual ward and see who represents them. Resolution status is public, so follow-up no longer depends on knowing the right office. Admins triage through a dedicated moderation workflow. Under the hood it's a Vite + React SPA on Supabase (Postgres with row-level-security policies and Edge Functions), deployed on Vercel, with ward-boundary data loaded from a GeoJSON dataset and rendered with MapLibre GL.

[03]/iters-live

PROD · LIVE

iters.live — All-in-One College Platform

The campus platform at SOA/ITER, used by about 12,000 students every week.

Live Deployment ↗GitHub ↗Case study →

// why: Campus life lived across dozens of WhatsApp groups — nothing central.

One campus hub: events, attendance, timetable, notes, PYQs in one place
Fast at peak: Redis cut DB load ~70%, responses under 200ms
One codebase, 3 clients: web, Android, and Electron shipped together

Node.jsExpressFirebaseRedisWebSocketJWT/RBACDockerElectron

// why this project?

Students had no single place to go. Everything lived across dozens of WhatsApp groups and manual processes.

// what does it do?

College platform at SOA/ITER serving 12K+ students. One hub for events, attendance, timetable, notes, PYQs, and AI grade analytics — replacing scattered WhatsApp groups. Open source, with web, Android, and Electron clients.

note: Operated solo alongside coursework; on-call for 12K users teaches you caching fast.

// results

Adopted as the primary campus platform within one semester. Redis caching cut database load by about 70% (measured with pg_stat_statements), keeping responses under 200ms at peak. One codebase builds for web, native Android, and an Electron desktop app, with a GitHub Actions pipeline running Jest unit tests and Playwright E2E checks on every push. All of it is run and operated by one person alongside a full course load.

November 2024

branch:release/iters-live

Sole developer & maintainer · Full Stack

verified telemetry

●12K+ active users

●100K+ daily API requests

●99.9% uptime

links & access

iters.live github.com/MrinallSamal-byte/iters.live

[03]/iters-live

PROD · LIVE

iters.live — All-in-One College Platform

The campus platform at SOA/ITER, used by about 12,000 students every week.

Live Deployment ↗GitHub ↗Case study →

// why: Campus life lived across dozens of WhatsApp groups — nothing central.

One campus hub: events, attendance, timetable, notes, PYQs in one place
Fast at peak: Redis cut DB load ~70%, responses under 200ms
One codebase, 3 clients: web, Android, and Electron shipped together

Node.jsExpressFirebaseRedisWebSocketJWT/RBACDockerElectron

// why this project?

Students had no single place to go. Everything lived across dozens of WhatsApp groups and manual processes.

// what does it do?

note: Operated solo alongside coursework; on-call for 12K users teaches you caching fast.

// results

Adopted as the primary campus platform within one semester. Redis caching cut database load by about 70% (measured with pg_stat_statements), keeping responses under 200ms at peak. One codebase builds for web, native Android, and an Electron desktop app, with a GitHub Actions pipeline running Jest unit tests and Playwright E2E checks on every push. All of it is run and operated by one person alongside a full course load.

July 2026

branch:release/gig-income-proof

Sole developer · Full Stack

verified telemetry

SHA-256 tamper detection●

OCR plus per-platform parsers●

Public verify links flag post-issue edits●

links & access

portable-income-credit-proof-for-gi.vercel.app github.com/MrinallSamal-byte/Portable-income-credit-proof-for-gig-workers

[04]/gig-income-proof

PROD · LIVE

Portable Income Proof for Gig Workers

Verifiable income records for gig workers, from earnings screenshots to lender-ready PDFs.

Live Deployment ↗GitHub ↗Case study →

// why: Gig workers earn real income but can't prove it to lenders.

Screenshots → proof: OCR + per-platform parsers extract earnings
Tamper-evident: SHA-256-hashed PDF with public verify link
Human in loop: review step to fix OCR before issuance

Node.jsExpressSupabasetesseract.jsGroq LLMEJSVercel Serverless

// why this project?

India's gig workers are excluded from formal credit because delivery/rideshare platforms issue no portable, verifiable income record.

// what does it do?

Turns gig earnings screenshots and PDFs into a tamper-evident, SHA-256-hashed income report any lender can verify via public link — no app install. Built for Swiggy, Ola, Urban Co. workers.

note: OCR on real earnings screenshots is messy — per-platform parsers + manual review beat pure LLM.

// results

You upload earnings screenshots or PDFs, tesseract.js OCR and per-platform parsers extract the numbers, and an optional Groq LLM pass fills gaps before a review step where you correct anything it got wrong. The result lands on an aggregated dashboard and exports as a SHA-256-hashed, shareable PDF whose public verify link detects edits after issuance. Auth is Google OAuth plus email OTP, data sits in Supabase Postgres behind row-level security, and the node:test suite passes on real OCR fixtures. Account Aggregator Phase-2 is stubbed.

[04]/gig-income-proof

PROD · LIVE

Portable Income Proof for Gig Workers

Verifiable income records for gig workers, from earnings screenshots to lender-ready PDFs.

Live Deployment ↗GitHub ↗Case study →

// why: Gig workers earn real income but can't prove it to lenders.

Screenshots → proof: OCR + per-platform parsers extract earnings
Tamper-evident: SHA-256-hashed PDF with public verify link
Human in loop: review step to fix OCR before issuance

Node.jsExpressSupabasetesseract.jsGroq LLMEJSVercel Serverless

// why this project?

India's gig workers are excluded from formal credit because delivery/rideshare platforms issue no portable, verifiable income record.

// what does it do?

Turns gig earnings screenshots and PDFs into a tamper-evident, SHA-256-hashed income report any lender can verify via public link — no app install. Built for Swiggy, Ola, Urban Co. workers.

note: OCR on real earnings screenshots is messy — per-platform parsers + manual review beat pure LLM.

// results

You upload earnings screenshots or PDFs, tesseract.js OCR and per-platform parsers extract the numbers, and an optional Groq LLM pass fills gaps before a review step where you correct anything it got wrong. The result lands on an aggregated dashboard and exports as a SHA-256-hashed, shareable PDF whose public verify link detects edits after issuance. Auth is Google OAuth plus email OTP, data sits in Supabase Postgres behind row-level security, and the node:test suite passes on real OCR fixtures. Account Aggregator Phase-2 is stubbed.

[05]/bhoomisahayak

SHIPPED

BhoomiSahayak — Land Record Correction Assistant

Finds gaps in Indian land records and generates a ready-to-file correction packet.

// why: Record errors block loans and inheritance; portals are opaque to navigate.

Live portal automation: walks KA District→Survey cascade on the real site
Gujarat PDFs parsed: privacy-safe, files deleted right after parse
Ready-to-file: printable mutation packet, manual fallback included

Node.jsReactVitePlaywrightSQLite

// why this project?

Land-record errors block loans, sales, and inheritance for millions of Indians. Finding out what's wrong means navigating opaque government portals first.

// what does it do?

Finds gaps in land records and generates a ready-to-file correction packet. Automates the live Karnataka Bhoomi portal via Playwright, parses Gujarat AnyROR PDFs, prints a mutation packet.

note: Started as blockchain, killed it after field research — the gap was workflow, not integrity.

// results

Automates the real Karnataka RTC portal end to end, walking its District, Taluk, Hobli, Village, and Survey cascade, verified against the production site. Parses Gujarat AnyROR PDFs privacy-safely, deleting uploaded files right after parsing, with a manual-entry fallback and printable application packets. It grew out of an earlier blockchain concept after field research showed the real gap is workflow, not data integrity.

July 2026

branch:release/bhoomisahayak

Sole developer · Civic Tech

verified telemetry

●Live Karnataka portal automation

●Gujarat AnyROR PDF parsing

●Uploaded files deleted after parse

[05]/bhoomisahayak

SHIPPED

BhoomiSahayak — Land Record Correction Assistant

Finds gaps in Indian land records and generates a ready-to-file correction packet.

// why: Record errors block loans and inheritance; portals are opaque to navigate.

Live portal automation: walks KA District→Survey cascade on the real site
Gujarat PDFs parsed: privacy-safe, files deleted right after parse
Ready-to-file: printable mutation packet, manual fallback included

Node.jsReactVitePlaywrightSQLite

// why this project?

Land-record errors block loans, sales, and inheritance for millions of Indians. Finding out what's wrong means navigating opaque government portals first.

// what does it do?

Finds gaps in land records and generates a ready-to-file correction packet. Automates the live Karnataka Bhoomi portal via Playwright, parses Gujarat AnyROR PDFs, prints a mutation packet.

note: Started as blockchain, killed it after field research — the gap was workflow, not integrity.

// results

Automates the real Karnataka RTC portal end to end, walking its District, Taluk, Hobli, Village, and Survey cascade, verified against the production site. Parses Gujarat AnyROR PDFs privacy-safely, deleting uploaded files right after parsing, with a manual-entry fallback and printable application packets. It grew out of an earlier blockchain concept after field research showed the real gap is workflow, not data integrity.

August 2026

branch:release/aapdasetu

Sole developer & architect · Disaster Response

verified telemetry

Zero-auth one-tap SOS●

Explainable RED/YELLOW/GREEN triage●

Routing avoids live flood polygons●

links & access

aapdasetu-v3.vercel.app github.com/MrinallSamal-byte/SIH

[06]/aapdasetu

LIVE · OPEN SOURCE

AapdaSetu — Disaster Response Platform

Zero-auth SOS, explainable triage, flood-aware GIS routing, and offline mesh comms for cyclone response.

Live Deployment ↗GitHub ↗Case study →

// why: In cyclones comms collapse — victims need help without accounts.

One-tap SOS: zero-auth, explainable RED/YELLOW/GREEN triage
Flood-aware routing: Sentinel-1 polygons fed straight into OSRM
Works dead-zone: offline PWA outbox + BLE mesh clients

Next.jsReact 19LeafletExpressPrismaFastAPISupabaseSwiftKotlin

// why this project?

In cyclones and floods, communication channels collapse and control rooms overload. Victims need one-tap help without creating accounts, and coordinators need triage they can actually trust.

// what does it do?

SIH disaster platform: zero-auth SOS, explainable RED/YELLOW/GREEN triage, flood-aware OSRM routing, Sentinel-1 flood maps, offline BLE mesh. Monorepo with command center, GIS, API, AI engine, mobile clients.

note: Hardest integration: SAR flood polygons → routing. Otsu thresholding finally behaved.

// results

Data lives in a ten-table Supabase schema with row-level security and realtime sync. Around 40 admin routes sit behind JWT auth, with Twilio and WhatsApp broadcast as fallback channels. An offline-first PWA queues reports in an IndexedDB outbox (bridged by BroadcastChannel) whenever connectivity drops. Flood polygons come from a Sentinel-1 SAR pipeline with Otsu thresholding and feed straight into evacuation routing. On top of that: pHash anti-fraud for damage photos, a psychological-first-aid chatbot module, and native BLE mesh clients for dead zones. Live v3 on Vercel, fully open source.

[06]/aapdasetu

LIVE · OPEN SOURCE

AapdaSetu — Disaster Response Platform

Zero-auth SOS, explainable triage, flood-aware GIS routing, and offline mesh comms for cyclone response.

Live Deployment ↗GitHub ↗Case study →

// why: In cyclones comms collapse — victims need help without accounts.

One-tap SOS: zero-auth, explainable RED/YELLOW/GREEN triage
Flood-aware routing: Sentinel-1 polygons fed straight into OSRM
Works dead-zone: offline PWA outbox + BLE mesh clients

Next.jsReact 19LeafletExpressPrismaFastAPISupabaseSwiftKotlin

// why this project?

In cyclones and floods, communication channels collapse and control rooms overload. Victims need one-tap help without creating accounts, and coordinators need triage they can actually trust.

// what does it do?

note: Hardest integration: SAR flood polygons → routing. Otsu thresholding finally behaved.

// results

Data lives in a ten-table Supabase schema with row-level security and realtime sync. Around 40 admin routes sit behind JWT auth, with Twilio and WhatsApp broadcast as fallback channels. An offline-first PWA queues reports in an IndexedDB outbox (bridged by BroadcastChannel) whenever connectivity drops. Flood polygons come from a Sentinel-1 SAR pipeline with Otsu thresholding and feed straight into evacuation routing. On top of that: pHash anti-fraud for damage photos, a psychological-first-aid chatbot module, and native BLE mesh clients for dead zones. Live v3 on Vercel, fully open source.

[07]/janseva-ai

SHIPPED

JanSeva AI — Welfare Scheme Intelligence

Enter a family's details once and it works out which of 26 welfare schemes they qualify for.

// why: Eligible families miss benefits because criteria hide across portals.

One profile, 26 schemes: deterministic engine, never LLM-invented
Live layer: deadlines, documents, news via 5 search agents
740+ districts: what-if simulator, vault, tracker, EN/HI voice

Next.js 16React 19PrismaSQLiteZustandTailwind 4z-ai SDK

// why this project?

Eligible families miss out on benefits they're entitled to because criteria are fragmented across opaque portals and nobody can tell them what applies.

// what does it do?

Enter a family's details once; a deterministic rule engine scores 26 central/state schemes. A 5-agent live-search layer adds deadlines, documents, news, and re-ranking. Bilingual chatbot, voice, appeal letters, certificates.

note: Kept entitlements deterministic on purpose — LLMs only add deadlines/docs, never decide.

// results

A 408-line deterministic eligibility engine keeps entitlement decisions auditable, so qualifications are never invented by a model. The LLM layer only adds deadlines, documents, and news. There's an SSE-streamed staggered agent pipeline underneath, district data covering 740+ Indian districts, plus a what-if simulator, document vault, application tracker, and guided quiz.

June 2026

branch:release/janseva-ai

Sole developer · Civic Tech

verified telemetry

●26 schemes scored by rule engine

●740+ districts covered

●408-line deterministic eligibility engine

[07]/janseva-ai

SHIPPED

JanSeva AI — Welfare Scheme Intelligence

Enter a family's details once and it works out which of 26 welfare schemes they qualify for.

// why: Eligible families miss benefits because criteria hide across portals.

One profile, 26 schemes: deterministic engine, never LLM-invented
Live layer: deadlines, documents, news via 5 search agents
740+ districts: what-if simulator, vault, tracker, EN/HI voice

Next.js 16React 19PrismaSQLiteZustandTailwind 4z-ai SDK

// why this project?

Eligible families miss out on benefits they're entitled to because criteria are fragmented across opaque portals and nobody can tell them what applies.

// what does it do?

note: Kept entitlements deterministic on purpose — LLMs only add deadlines/docs, never decide.

// results

A 408-line deterministic eligibility engine keeps entitlement decisions auditable, so qualifications are never invented by a model. The LLM layer only adds deadlines, documents, and news. There's an SSE-streamed staggered agent pipeline underneath, district data covering 740+ Indian districts, plus a what-if simulator, document vault, application tracker, and guided quiz.

initial commit · 2024.01

$ cat resume.md

Download Resume

One page, updated Aug 2026. Skim the core credentials below or grab the PDF directly.

●Mrinall_Samal_Resume.pdf

Format: PDFUpdated: Aug 2026Target: SWE / AI-ML

// executive summary & verified credentials

▸

Academic Excellence: CGPA 9.54 / 10.0 in B.Tech CSE at SOA/ITER University.

▸

Wipro AI/ML Internship: Built & optimized distributed training pipelines with INT8 quantization for Fortune 500 clients (Mar–Sep 2025).

▸

National Hackathon Placements: Top 50 in Smart India Hackathon (SIH) 2026, EY Techathon finalist (500+ teams), MumbaiHacks finalist, Adobe finalist, 6th nationally in Geo AI (IIT Bombay).

▸

Google Cloud Platform: Diamond Rank (top 1% globally) on Google Cloud Skills Boost.

▸

Live Production Systems: iters.live serving ~12,000 students and ~100K requests/day; CodeMentor AI interactive developer tutor; Amma Odisha civic portal live with ward GIS mapping.

Download Resume (PDF)LinkedIn

// contact