Soumyadeb Tripathy

PROFILE

Computer science engineer focused on software that stays correct under concurrency, partial failure, and adversarial input — with claims that survive independent verification. Shipped 23 public repositories in ten months, from a Linux kernel I/O scheduler and a Rust cross-language analysis engine to audited research platforms; published two null results rather than reframing them.

SELECTED SYSTEMS

• AdityaNet — Astro 7, TypeScript (strictest), React 19 islands, Python · A research platform where no number is typed by a person. Key result: ml vs. threshold detector no gain.
• Cartograph — Rust, tree-sitter, petgraph, TypeScript/TSX + Python analysis · Architecture graphs that show their evidence, not their guesses. Key result: cross-stack chain resolves TS → Python.
• FurnitureOps — Next.js 14, TypeScript, Supabase Postgres, Upstash Redis · One invariant, defended end to end. Key result: verification scripts 19.
• GEOFENCE-LLM — Python, PyTorch, Transformers, scikit-learn · Jailbreak detection that attackers can't paraphrase around. Key result: obfuscated-prompt recall 0.66.
• Nexus-RTB — Python, LightGBM, Prometheus, Grafana · Every microsecond and every unit of budget accounted for. Key result: p99 request path 0.15ms.
• ASTRA — Python, PyTorch, lightkurve, ONNX · Classifying variable stars with a pipeline that audits itself. Key result: test accuracy 78.17%.
• DDSO — C, Linux kernel, ftrace, Node.js · An I/O scheduler that rewrites its own strategy. Key result: scheduling strategies 3.
• Helios-Dx — PyTorch, PennyLane, Concrete-ML, Docker · The experiment that said no. Key result: quantum advantage observed none.

CAPABILITIES

• BACKEND SYSTEMS — Node.js, FastAPI, Next.js API routes, REST
• DISTRIBUTED COMPUTING — async queue architectures, row-level locking, PID pacing, budget coordination
• APPLIED MACHINE LEARNING — PyTorch, LightGBM/XGBoost, CNN+Transformer hybrids, knowledge distillation
• RESEARCH ENGINEERING — reproducible pipelines, digest-addressed datasets, build-time evidence gates, ground-truth audits
• SYSTEMS & PROGRAM ANALYSIS — Rust, tree-sitter, static analysis (symbol extraction, route canonicalisation), C
• SECURITY & AI SAFETY — LLM jailbreak defense, hidden-state interpretability, fail-closed control loops, CSP/HSTS hardening
• PRIVACY ENGINEERING — fully homomorphic encryption (Concrete-ML, fhEVM), client-side AES-256-GCM, on-chain access control, local-first inference
• FRONTEND SYSTEMS — React 19, Next.js 14–16, TypeScript, Tailwind CSS

EDUCATION

B.Tech, Computer Science and Engineering — Lovely Professional University (2025 – present)

Higher secondary — Kishorenagar Sachindra Siksha Sadan (2024)

LANGUAGES

English · Hindi · Bengali