LAKSHYA
I design, build, and operate the full stack myself — from bare-metal Linux and self-hosted services to GPU inference servers and privacy-first browsers. Data stays on hardware I control.
RESEARCHER.
ENGINEER.
SYSTEM BUILDER.
I am a Computer Science undergraduate at JK Lakshmipat University building privacy-first desktop applications, local-first mobile apps, security tooling, and clinical ML pipelines. My work spans Electron browsers with renderer isolation, C++20 DNS security agents, Flutter finance trackers with 159 tests, and MIMIC-IV ICU early-sepsis prediction with XGBoost (AUROC 0.7935).
My engineering philosophy: ship tested, offline-capable software that respects user privacy. Languages: Python, JavaScript, C++, Dart, SQL. Certified: CITI Program — Data or Specimens Only Research (Score 97).
AI & Clinical ML
「 医療機械学習 」MIMIC-IV ICU Sepsis-3 temporal pipelines, LSTM sequence modeling (AUROC 0.779), and GroupKFold cross-validation.
Local LLM & Agents
「 自律型エージェント 」Dedicated RTX 4070 model server with Ollama, Nginx reverse proxy, and isolated WSL2 Hermes Agent profiles.
Linux & Systems
「 基盤設計 」Debian/Ubuntu production workflows, zero public ports via Cloudflare Tunnels, Tailscale zero-trust, and network isolation.
Docker & Homelab
「 自作環境 」Self-hosted media, storage, automated Git/Docker Compose deployment pipelines, and persistent service stacks.
THE ARSENAL
Battle-tested tools, frameworks, and low-overhead runtimes powering reproducible ML models and resilient infrastructure.
THE CHRONICLES
Documented appointments, clinical milestones, and academic track records that prove capability in the field.
Undergraduate Research Intern — Early Sepsis Prediction
MIMIC-IV | Supervised by Dr. Amit Sinhal, JKLUBuilt an end-to-end ML pipeline for 6-hour-horizon Sepsis-3 prediction on MIMIC-IV ICU data: suspected-infection identification, patient/admission/ICU preprocessing, and temporal feature engineering. Benchmarked five approaches (XGBoost, Random Forest, LSTM, Logistic Regression, qSOFA) on identical windows with 5-fold GroupKFold splits by patient to prevent leakage. XGBoost led at AUROC 0.7935 / AUPRC 0.0275 at 0.60% window prevalence; applied SHAP for interpretability. External validation on eICU underway.
Freelance Tech & AI Writer
MediumPublishing detailed tutorials and architectural deep-dives on Linux virtualization, lightweight homelab containers, and local AI workflows. Notable pieces include 'I Ran 4 Operating Systems on a 4 GB Home Server' and '30 Days Replacing Windows with Linux'.
B.Tech in Computer Science
JK Lakshmipat University — Jaipur, IndiaSelected for the R-CAT Quiz-A-Thon programme. Relevant coursework: Machine Learning, Data Structures & Algorithms, Computer Networks, Database Management Systems, Cloud Computing, and System Design.
SELECT VOLUMES
Systems, machine learning architectures, and applications crafted with production rigor.
Clinical Early Sepsis Prediction
End-to-end ML pipeline for 6-hour-horizon Sepsis-3 prediction on MIMIC-IV ICU data. Benchmarked XGBoost, Random Forest, LSTM, Logistic Regression, and qSOFA with strict 5-fold GroupKFold patient-level splits. XGBoost led at AUROC 0.7935 / AUPRC 0.0275; SHAP used for interpretability.
Aegis Privacy Browser
Privacy-focused Electron browser with renderer isolation via preload bridge, nodeIntegration disabled across all windows. Features ~1,400 ad-block/anti-fingerprinting filters across 3 aggressiveness levels, download manager, tab suspension, incognito sessions, and a 13-test Node.js built-in suite guarding a false-positive URL corpus.
SentinelDNS
Local-first DNS security agent co-built in C++20. Core handles wire encode/decode, TTL-bound cache, policy matching, UDP listener/resolver, and a telemetry sink. Policy is a reviewed, git-tracked file with no API write path; diagnostics exposed via a read-only NDJSON IPC API.
CashWand Finance Tracker
Offline-first Flutter finance tracker with SQLite persistence, overall and category budgets with real-time over-budget warnings, and biometric app lock (fingerprint, Face Unlock, PIN). Ships with 159 tests across entities, repositories, and providers, plus a manual UAT plan.
CREDENTIALS
Official certifications, verified research compliance, and academic milestones.
CITI Program — Data or Specimens Only Research
CITI Program · Biomedical & Clinical Data Ethics
Verified certification covering ethical handling of human subject data, de-identification protocols, and research governance for clinical machine learning pipelines.
MIMIC-IV Sepsis Prediction Research
Supervised by Dr. Amit Sinhal · JKLU
Undergraduate research internship investigating 6-hour sepsis-3 prediction using patient-level GroupKFold validation, benchmarking XGBoost, Random Forest, and LSTM with AUROC 0.7935.
TECH WRITING
Operational logs, homelab benchmarks, and architectural tutorials published on Medium.
LET'S BUILD.
Open to AI/ML research collaborations, systems engineering roles, and resilient infrastructure setups. Reach out directly or dispatch a transmission.
© 2026 LAKSHYA AGARWAL. INKED MANGA PANEL EDITION.
ENGINEERED WITH NEXT.JS 16 • TAILWIND CSS • ZERO DEPENDENCIES