Open to internships (Jan – Jun 2027) and full-time roles
Rahul Paul
Computer Science (Cyber Security) undergraduate, Bengaluru, India.
I build where machine learning, backend systems and security meet, and I stay with a problem through the unglamorous parts: the tests, the CI, and the bug that quietly halves your scores.
$ cat about.md
About
Most of what I know, I learned by building. My projects sit where machine learning, backend engineering and security overlap: a reconciliation engine that has to explain every record it can't match, a DNS threat detector tuned against a real false-positive budget, and an on-device captioning model cut down to 146 MB.
I care about shipping things that actually run and about measuring them honestly, so most of my repos come with an evaluation harness and a CI check that fails when the numbers regress. I'm looking for software engineering, machine learning and security roles: internships from January to June 2027, and full-time positions after I graduate in 2027.
$ cat education.txt
MS Ramaiah Institute of Technology
B.E. Computer Science and Engineering (Cyber Security)
2023 – 2027 CGPA 9.52 / 10
Bhavan's Gangabux Kanoria Vidyamandir, Kolkata
Class XII (AISSCE)
2021 – 2022 96.33%
$ tail experience.log
Work
[Dec 2025 – Jul 2026]
Project Intern
Samsung PRISM, Samsung R&D Institute India
- Took EmoCapNet from research prototype to an on-device build. Distilled a ViT + GPT-2 captioner into a TinyCLIP-ViT-8M + DistilGPT-2 student, then used quantization-aware training for INT8: 6.8x smaller at 146 MB and about 450 ms per caption on CPU, with KV-cache decoding written from scratch and an ONNX/TorchScript export path.
- Designed the model to be steered by a continuous valence-arousal-dominance (VAD) vector instead of discrete emotion labels, using cross-attention fusion and VAD-FiLM conditioning trained multi-task on an emotion-augmented Flickr30k corpus. Continuous control halved directional error against a discrete-label baseline.
Teacher
ViT + GPT-2 captioner
Distil
soft-logit + feature losses
Student
TinyCLIP-ViT-8M + DistilGPT-2
INT8
quantization-aware training
On-device
146 MB, ~450 ms on CPU
$ ls -l projects/
Projects
$ ls ~/github --more
- Rubiks-Cube-SolverC++IDA* with pattern-database heuristics on a bitboard; solves scrambles in under a second.
- multi-source-candidate-transformerPythonMerges conflicting candidate data into one deduplicated profile with provenance on every field.
- AI-Botnet-Traffic-DetectionPythonXGBoost/LightGBM on live Scapy captures, with a FastAPI + Streamlit dashboard and SHAP.
- AI-Help-ExtensionJavaScriptA Gemini tutor inside a CP platform that reads the problem and your code to give hints.
$ git log --all --author=rahul
$ tree skills/
Skills
skills/
├── languages/
│ └── C / C++, Python, TypeScript, JavaScript, SQL, Bash
├── foundations/
│ └── Data structures & algorithms, Object-oriented design, Operating systems, Computer networks, DBMS
├── backend/
│ └── FastAPI, Node.js / Express, REST, SSE, SQLAlchemy, SQLite, MongoDB, React, Three.js
├── testing-and-ops/
│ └── pytest, Jest, Vitest, Playwright, autocannon, Mutation testing, Docker, GitHub Actions, CodeQL
└── ml-and-security/
└── PyTorch, ViT / GPT-2, Knowledge distillation, INT8 quantization, ONNX, scikit-learn, SHAP, LiteLLM, Sandboxed execution, Threat detection
$ ls ~/interests/
Beyond the code
Sport
- European football
- Cricket
- Tennis
The world
- Geopolitics
- Politics
- Current affairs
Intelligence
- Intelligence operations
- Espionage history
Humanities
- History
- Art
- Culture
Songs
My YouTube playlist, 377 tracks and counting.
$ mail rahul
Contact
I'm open to software engineering, machine learning, quant and security roles: internships from January to June 2027, and full-time positions after I graduate in 2027. Email is the fastest way to reach me.