Systems software engineer
Kshitij Shettigar.
Third-year CS undergrad at PES University, Bangalore. Building Parakaya, a Rust VM that migrates live processes between machines.
- gskshitij0638@gmail.com
- Bangalore, IST
- Open to summer-2027 internships
Focus areas: Rust, process VMs, supply-chain security, zero-knowledge crypto, FHIR, robotics.
Selected work
Parakaya
A running program hops between a PC, a Raspberry Pi, a phone, and a cloud VM mid-execution, across CPU architectures, surviving power loss at any instant.
Case studyPassGen Vault
A zero-knowledge password manager: one AES-256-GCM vault, opened by a web app and a native Android client.
Supply-Chain Trust Gate
Checks a registry's "fixed in version X" claim against the files PyPI actually shipped, before a pipeline installs the package.
In numbers
Open to summer-2027 internships.
gskshitij0638@gmail.comFive projects, five different problems.
Parakaya
Rust, custom instruction set, in active development
Problem
Make a running program hop between a PC, a Raspberry Pi, a phone and a cloud VM mid-execution, across CPU architectures, and survive power loss at any instant.
Approach
The VM core is a Rust workspace: a compact instruction set, tagged values, and execution state built to checkpoint on one architecture and resume on another. Designed spec-first, with decision records for every architectural call.
Result
In active development, so there are no results to claim yet. The instruction set below is from the current VM core.
pub enum Instr { I32Const(i32), I32Add, I32Sub, I32Mul, I32DivS, LocalGet(u32), LocalSet(u32), Block, Loop, If, Else, Br(u32), BrIf(u32), Return, Call(u32), I32Load(MemArg), I32Store(MemArg), End, }
Supply-Chain Trust Gate
Capstone, team of four, PES University. I own the PyPI adapter
Problem
Build pipelines and AI coding assistants act on what a vulnerability database says, such as "fixed in version X", without checking it against what was actually shipped. When GitHub and PyPA disagree, someone acts on a wrong record.
Approach
Fingerprint the advisory's fix commit, download every PyPI release of the package, and find the first release that really contains the fix. A trust router then answers TRUST, VERIFY, FLAG or ABSTAIN, with the evidence, before anything is installed.
Result
Stage 0, written down before we counted: 820 real cases where GitHub and PyPA name different fixed versions (pass mark 300), 711 with a fix commit we can trace. Phase I runs to November 2026.
PassGen Vault
Single-file web app and native Android client
Problem
One password vault shared by a laptop and an Android phone, without trusting any server with the plaintext.
Approach
One AES-256-GCM vault stored in OneDrive, opened by a single-file web app and a native Android client. Keys come from PBKDF2 at 600,000 iterations, and fingerprint unlock is backed by hardware keys.
Result
A working zero-knowledge password manager. Its generator is built on a CSPRNG with rejection sampling, so there is no modulo bias.
Clinical Timeline Builder
Medical Innovations and Technology minor, PES University
Problem
Represent a diabetic care pathway as standard FHIR R4 data, then turn that data back into something a clinician can read.
Approach
Twenty hand-authored FHIR R4 resources, coded with LOINC, ICD-10 and RxNorm and validated against a HAPI server.
Result
A Python pipeline that rebuilds the resources into a chronological clinical timeline.
Repo6uard
Team of six, CIE Ignite startup program at PES
Problem
Pattern-based scanners flag code without context and say nothing about vulnerable dependencies.
Approach
A hybrid scanner: static analysis finds known patterns, an LLM reasoning pass validates them in context and writes fix suggestions, and dependencies get cross-checked against Google's OSV and the NVD.
Result
The working prototype scanned a deliberately vulnerable repo and reported 15 issues across 8 OWASP categories in under 30 seconds. We pitched it through two phases of the campus startup program and missed the final round by one position. The scanner still works.
More on GitHub
Software engineering, end to end.
From requirements to release. I design before I code, test what I ship, automate the pipeline, and work with AI coding agents with every change reviewed.
Repo6uard
Full-stack web app, team of six, CIE Ignite startup program. I built the backend pipeline
Problem
Developers need vulnerability findings they can act on before they deploy, not a wall of pattern matches with no context.
Approach
A React 19 and Vite interface over a Flask API with scan, repo-scan and health endpoints. My part was the multi-pass backend: Bandit and njsscan static analysis, an LLM pass that drops false positives and writes fixes, and a live OSV dependency check.
Result
15 issues across 8 OWASP categories on a deliberately vulnerable repo in under 30 seconds. We pitched it through two phases of the startup program and missed the final round by one position.
Multi-Agent Development Environment
How I write software day to day
Problem
Two AI coding agents, Claude Code and Codex, each want their own rules, skills and memory. Keeping two copies in sync drifts, and unreviewed agent output is a liability.
Approach
One shared set of skills, project rules and read/write memory that both agents load, with a written design spec covering parity, permissions and sandboxing. Work runs planner, executor, reviewer: a lead agent writes bounded instructions, sub-agents change only what I approve.
Result
Every task, from coursework to the capstone, runs through one setup. The lead agent verifies each change before it is reported done, so I review evidence, not claims.
Agentic Study Pipeline
Course material in, knowledge graphs and practice papers out
Problem
Slides, notes and question banks for five courses sit behind a university portal, and a downloader that only half works fails silently.
Approach
A sync tool pulls every course's material. A per-subject knowledge graph stores theorems, worked problems and past exam questions as typed nodes, each tagged with its common exam trap. An AI agent retrieves from the graph to write practice papers and run quiz drills.
Result
Found and fixed a bug where only 21 of 62 classes' files were downloading. The Graph Theory graph holds 409 nodes and 543 edges, and a slide-diff check keeps the agent on verified sources.
Meditech Lab
Healthcare informatics dashboard, Medical Informatics course
Problem
Simulate a hospital lab pipeline end to end: register a patient, record results, and emit the standard messages real hospital systems exchange.
Approach
A modular Flask app over PostgreSQL, with a pure domain layer kept apart from HTTP and database wiring so it tests without a database. An HL7 v2.5 message builder written from scratch, running on a five-service Docker Compose stack.
Result
About 3,300 lines of Python across 26 commits and five modules: patient registry, intake, a clinical-note entity extractor, an ICD-10 and SNOMED coder, and a live pipeline view.
How I build
Toolkit
Languages
- Python, Java
- C, C++, JavaScript
- SQL, Rust
Web
- React, Vite
- Flask, REST APIs
- HTML, CSS
Databases
- PostgreSQL
- MongoDB
- SQLite
AI & agents
- Claude Code, Codex
- RAG, prompt engineering
- MCP tools, LLM APIs
Engineering practice
- OOP, data structures
- Agile, Scrum
- Testing, code review
Delivery
- Git, GitHub Actions
- Docker Compose
- Linux
Journey
- Now
CS at PES University, with a minor in Medical Innovations and Technology.
Capstone: Supply-Chain Trust Gate. Electives this semester: AR/VR (OpenGL, Unity, Three.js) and Graph Theory.
- 2026
Completed a two-month IoT and Robotics internship at Tripple One Solutions (Jun-Aug).
Embedded firmware, motor control and device-to-cloud connectivity.
- 2026
Built a heart-disease prediction model at the Epoch Datathon.
- 2022
Ranked 5th in school, CBSE Class 10 board examinations.
- 2019
Represented India at the FIRST LEGO League International Championship in Sydney.
- 2019
1st place and Best Project Award, FIRST LEGO League Regional.
About
I got into computers through robotics: FIRST LEGO League, then representing India at the 2019 International Championship in Sydney. These days the obsession is lower level. I am an Arch Linux and Hyprland person, and I like problems where the OS, the hardware and the network all have opinions.
At PES University I study CS with a minor in Medical Innovations and Technology, which is where the FHIR work comes from. Over the summer of 2026 I did a two-month IoT and Robotics internship at Tripple One Solutions, working on embedded firmware, motor control and device-to-cloud connectivity. My capstone is a supply-chain trust gate that checks vulnerability records against the packages actually shipped, and I am ramping up on Kubernetes, Ansible and AWS next.
Outside of code: keyboard, music production and cycling.
Languages
- Python, C, C++
- Rust
- Java, JavaScript
Systems & DevOps
- Linux (Arch, Hyprland)
- Docker, Git, GitHub Actions
- Arduino, Raspberry Pi
Security
- Supply-chain security
- Web Crypto, AES-GCM
- Wireshark, Bandit, OSV
Web
- React, Node, Express
- Flask, REST APIs
- MongoDB, SQLite
Open to internships and collaboration.
If you are working on systems, security or health-tech and want an engineer who reads the spec first, get in touch.
- GitHub
- github.com/Kishi836
- Based in
- Bangalore, India (IST)