Software Engineer

Shashwat Manimaran
Building across systems, security and full-stack.

Real-time C++ pipelines, raw-disk forensics tools, and secure products deployed on AWS.

View Projects
Get in touch
Shashwat Manimaran
About me

I'm an Integrated M.Tech (Software Engineering) student at VIT. As an intern at HCLTech I built a real-time C++20 weather data pipeline and a low-latency Linux alerting service.

My own projects range from a raw-disk NTFS forensics toolkit written in pure Python, to a full-stack expense platform with a native Android client deployed on AWS, to an offline learning platform for schools without internet that runs on a Raspberry Pi. I'm most interested in systems software, semiconductors and security.

  • 9.45 CGPA · Integrated M.Tech, Software Engineering, VIT
  • 2 Published Indian patent applications in AI governance and LLM security
  • AWS Certified Cloud Practitioner
  • HCLTech Software Engineering Intern, 2026
Tech stack
Skills

What I build with

Most of my work sits close to the system: concurrency, on-disk formats and the security around them. I carry it through to tested, deployed products.

01

Languages

  • C
  • C++ (C++20)
  • Python
  • Go
  • Kotlin
  • TypeScript
  • JavaScript
  • Java
  • SQL
02

Systems & low-level

  • Linux system programming
  • Embedded Linux (Raspberry Pi)
  • Multithreading
  • Lock-free data structures
  • Real-time scheduling (SCHED_FIFO)
  • On-disk format parsing (NTFS)
  • Cross-compilation
  • gdb
03

Security

  • Digital forensics ($MFT, USN journal, ADS)
  • Timestomping detection
  • Auth hardening
  • Rate limiting & CSP
  • HMAC
  • AES-GCM / Android Keystore
  • LLM prompt-injection defence
04

Cloud & DevOps

  • AWS (ECS Fargate, ALB, IAM, Secrets Manager, CloudWatch)
  • Docker
  • Docker Compose
  • Terraform
  • GitHub Actions
  • systemd
  • Vercel
05

Web & mobile

  • React
  • Next.js
  • Node.js
  • TanStack Query
  • Tailwind CSS
  • Auth.js
  • REST APIs
  • Android (Jetpack Compose, Hilt, Retrofit)
06

Data

  • TimescaleDB / PostgreSQL
  • MongoDB
  • SQLite
  • Redis
  • IndexedDB
07

Observability

  • Prometheus
  • Grafana
  • Sentry
  • CloudWatch
08

AI / ML

  • llama.cpp (on-device inference)
  • OpenCV
  • OpenAI-compatible LLM APIs
09

Testing & tools

  • pytest
  • Vitest
  • Paparazzi
  • MockWebServer
  • Contract testing
  • Git
  • PyInstaller
From raw disk sectors
to production
Experience

Where I've worked

HCLTech

Software Engineering Intern · Chennai, India

May — Jul 2026
PROJECT

Real-time weather data pipeline & alerting service

A C++20 pipeline that ingests live weather data from the Open-Meteo API into TimescaleDB in three stages (multithreaded collectors, hierarchical aggregators and a batching writer), plus a low-latency Linux alerting service that compares live readings against hourly baselines.

Fault-tolerant storage layer: transactional writes in batches of 250 rows, duplicate removal, and a file-based outbox that replays buffered data after database outages
Memory stays bounded under overload: queues are capped at 4,096 messages, and backpressure events are exported to Prometheus and shown in Grafana
Real-time alerting service built on a lock-free ring buffer with SCHED_FIFO scheduling
Two front ends: a Leaflet map console showing weather hazards and per-segment delays on freight corridors, and a Chart.js dashboard for historical analysis and CSV dataset comparison
Deployed with Docker Compose and systemd, and wrote the architecture and setup documentation
C++20
Multithreading
Lock-free
Linux SCHED_FIFO
TimescaleDB
Prometheus
Grafana
Leaflet
Chart.js
Docker Compose
systemd
Projects

Things I've built

A disk forensics toolkit, a deployed fintech platform and an offline classroom lab, each taken from idea to tested, working software.

01 Sep 2025 — Present

ForensicaLite

NTFS disk forensics toolkit

A raw-disk NTFS forensic analyzer in pure Python. It recovers deleted files, detects timestomping and rebuilds file history, without touching the OS file APIs.

  • 100,159MFT records parsed in 5.7 s on a 57 GB volume
  • 300K+USN journal records processed
  • 0Scan errors; layout matched Windows fsutil
  • 156Automated tests (pytest + Vitest)
How it works
  • Opens Windows volumes at the sector level (\\.\X:), bypassing OS file APIs and locks, and parses the boot sector, $MFT, $Bitmap and USN change journal by hand, using only the standard library.
  • The MFT parser handles update-sequence fixups, signed-delta data runs, fragmented $MFT extents and $ATTRIBUTE_LIST records. A 1 MiB chunk cache cuts physical reads to one per 1,024 records.
  • Recovers resident deleted files byte for byte, and grades how recoverable non-resident ones are by checking $Bitmap for reallocated clusters.
  • Flags timestomping (comparing $STANDARD_INFORMATION with $FILE_NAME, plus zeroed sub-second precision), hidden Alternate Data Streams and orphaned records, and rebuilds rename, move and delete history from the USN journal.
  • An offline React report app (triage, timeline, hex viewer, IndexedDB case storage, HTML/CSV export) means evidence never leaves the examiner's machine. A versioned JSON contract, checked in CI against real analyzer output, connects the two.
  • Ships as a tkinter GUI and a CLI packaged as standalone Windows executables. Tests run against a synthetic NTFS image built byte by byte, so they need no real drive or admin rights.
  • Python
  • TypeScript
  • React
  • Tailwind CSS
  • Vite
  • IndexedDB
  • JSON Schema
  • tkinter
  • PyInstaller
  • pytest
  • Vitest
GitHub Report app
02 Web · Android · AWS

FinTrack

Shared expense & settlement platform

A full-stack expense-splitting platform with escrow-style settlements, hardened authentication and a native Android client, deployed on AWS.

  • 733Automated tests on the web app
  • 5 → 0Client-side API calls on dashboard first load
  • ~12kLines of Kotlin across 12 screens
  • 28Paparazzi screenshot tests
How it works
  • A greedy debt-simplification algorithm with integer-paise maths, so totals never drift from rounding.
  • Escrow-style settlements: a logged payment only moves balances once the recipient confirms it. The server enforces this with conditional updates, so a double-submit can't bring back a declined payment.
  • Hardened Auth.js: bcrypt, password reset that stores only token hashes and logs out every session, responses that don't reveal whether an account exists, open-redirect-safe ?next=, Redis rate limiting with a per-process fallback, and CSP headers.
  • A daily recurring-expense job handles month-end dates, catches up on missed runs and claims each rule atomically, so two runs can't double-charge a group.
  • Containerised and deployed on AWS ECS Fargate behind an Application Load Balancer, with Terraform infrastructure, Secrets Manager, least-privilege IAM, CloudWatch alarms and a GitHub Actions pipeline.
  • Native Android client in Kotlin and Jetpack Compose. It uses Bearer-token sessions encrypted with AES-GCM in the Android Keystore, Google sign-in through Credential Manager, a cached bootstrap so it opens with real data, and invite deep links that survive sign-in.
  • Next.js
  • React
  • TypeScript
  • MongoDB
  • Redis
  • Auth.js
  • AWS ECS Fargate
  • Terraform
  • GitHub Actions
  • Kotlin
  • Jetpack Compose
  • Hilt
Web repo Android repo
03 Offline-first · Raspberry Pi

Tessera Quest

Offline gamified coding & STEM platform

Turns a room of computers with no internet into a working coding lab, using a Raspberry Pi, USB drives, printed cards and an on-device AI tutor.

  • 25Levels, each automatically checked to be solvable
  • 14Printed ArUco programming cards
  • 4Subjects: coding, chemistry, physics, math
  • 0Accounts, cloud services or network needed
How it works
  • Kids arrange printed ArUco cards on a desk, read by a Python/OpenCV camera pipeline, or drag matching Blockly blocks. Both compile to one JSON AST that runs on a deterministic Go executor.
  • 25 levels across sequencing, loops, nested loops, if/else, while and composition. Each one is run against the real executor to confirm it can be solved.
  • An on-device pet tutor runs Qwen3 0.6B or 1.7B through llama.cpp, choosing the size from available RAM. It only rewrites pre-verified, human-written hints and never judges code, which keeps a small model safe in front of children.
  • One cross-compiled Go binary serves the React frontend and JSON API, and owns a portable SQLite save file on each student's drive.
  • An optional classroom hub on the Pi collects progress for a teacher dashboard and restores lost drives. Sync is HMAC-authenticated, and a restore can only raise saved progress, never lower it.
  • Drives move between unmanaged machines, so a checksum manifest detects tampering at launch and the server listens only on loopback by default.
  • Go
  • React
  • TypeScript
  • Blockly
  • Python
  • OpenCV
  • SQLite
  • llama.cpp
  • Raspberry Pi 5
More on GitHub
Recognition

Patents, certifications & leadership

  • Patent

    A System and Method for Centralized AI Governance and Orchestration

    Co-inventor · Indian Patent Application IN202641075844 A1

    Published Jun 2026
  • Patent

    Prompt-Flow DAG Trust Graph for Multi-Hop Prompt Injection Detection in Multi-Agent LLM Workflows

    Co-inventor · Published Indian patent application

    Published
  • Certification

    AWS Certified Cloud Practitioner

    Amazon Web Services

    2026
  • Certification

    AWS Certified Solutions Architect – Associate

    In progress

    Expected Dec 2026
  • Leadership

    Treasurer, Toastmasters Club (Fogo)

    Managed club finances and coordinated meetings.

    Oct 2024 — Oct 2025