Sampark Bhol

Software Engineer · Helmit · Munich, Germany / remote from India · since Aug 2026

Backend systems · ML · LLM agents · games · mobile

Experience

Helmit — Software Engineer

Aug 2026 – present · Munich, Germany / remote, India · current

Design, develop, test and maintain applications and microservices using Java and Spring Boot, Go and TypeScript on AWS and GCP, owning REST API contracts, request validation, asynchronous processing, idempotent operations and reliable state transitions across a distributed application design serving iOS and Android clients.
Support application performance, reliability, resilience and operational readiness with structured logging, service metrics, health checks, timeout propagation, retry policies, backpressure and graceful degradation, and resolve production defects across application, service, network and relational database layers using latency, error-rate and saturation signals.
Contribute to CI/CD pipelines and automated deployments with GitHub Actions and Cloud Build pre-deploy gates covering build, vet, test and dependency freshness, applying DevSecOps practices through Secret Manager-backed configuration, least-privilege IAM, API gateway routing and WAF edge policies to keep services secure, scalable and maintainable.
Operate containerized Kubernetes (GKE Autopilot) workloads with horizontal pod autoscaling, disruption budgets and a leader-election sharding pattern that lets replicas divide ownership of live platform connections.
Backed by Redis job queues and PostgreSQL across 58 schema migrations.
Delivered 9 platform integrations end to end across OAuth2, MTProto and push-model data portability flows, including an undocumented third-party API reverse-engineered from its rejection responses and stabilised with a retry-and-drop strategy.
Built a tool-calling LLM agent on Vertex AI with an argument-resolved permission model, OpenTelemetry distributed tracing across the router, skill-selection and tool-execution loop, and a statically enforced data-quarantine boundary keeping untrusted content away from tool execution.
Contribute to the React Native and Expo mobile client with native iOS Screen Time and Family Controls integrations through custom Swift bridges, alongside Sentry, PostHog, Firebase and attribution instrumentation and en/de localization built as infrastructure rather than hardcoded strings across a live chat UI.

Stack: Java, Spring Boot, Go, TypeScript, Python, gRPC, AWS, GCP, GKE Autopilot, Redis, PostgreSQL, Vertex AI, OpenTelemetry, React Native, Expo, Swift

CLR3 Ventures (Spizen Technologies) — Founding Machine Learning Engineer 1

Jan 2026 – Aug 2026 · Bengaluru, India

Built a profitability prediction system with ensemble gradient boosting over event-driven Kafka streaming feature pipelines, reaching 92% accuracy on revenue forecasts and powering dynamic pricing across business verticals.
Shipped a transformer embedding similarity engine with approximate nearest neighbour search on ClickHouse and a neural collaborative filtering recommender serving 10K+ queries per second in production behind REST APIs.
Architected an autonomous content and synthetic data generation pipeline orchestrating multi-stage LLM agents through planning, drafting, summarization and validation, with retrieval-based deduplication and policy-aware filtering.
Established the founding ML infrastructure on Apache Kafka for event-driven streaming, with Redis caching, SQL feature stores, Grafana dashboards for observability, model drift detection and inference latency, plus GitHub Actions CI/CD pipelines with A/B testing for safe model rollouts.

Stack: Kafka, ClickHouse, Redis, Grafana, gradient boosting, transformers, LLM agents, GitHub Actions

QuantumX (Appetite Studio) — Quantum Computing & Research Intern

Nov 2025 – Jan 2026 · Remote

Researched hybrid quantum-classical models pairing variational quantum circuits with neural networks for combinatorial optimization, designing QAOA and VQE protocols and evaluating post-quantum cryptography, and authored quark, an open-source library learning semantic embeddings over QASM circuits on Neon PostgreSQL and RabbitMQ; first-authored a journal paper on quantum-enhanced machine learning, now under review.

Stack: QAOA, VQE, variational circuits, Python, QASM, Neon PostgreSQL, RabbitMQ

CSM Technologies — AI & ML Engineer Intern

May 2025 – Jul 2025 · Odisha, India

Shipped production RAG pipelines with LangChain, LangFlow, ChromaDB and Pinecone, lifting domain accuracy by 25% through chunking and prompt tuning; built BERT and GPT-class transformer applications; fine-tuned LLMs with LoRA and QLoRA at 60% less training compute, containerized with Docker and deployed on Kubernetes.

Stack: LangChain, RAG, LoRA/QLoRA, Docker, Kubernetes

National Informatics Centre, MeitY (Government of India) — Deep Learning & Software Development Intern

May 2024 – Jul 2024 · Odisha, India

Built anomaly detection with Isolation Forest, DBSCAN, z-score statistics and graph analysis on 1M+ government records at 95% precision-recall; predictive models with full EDA and feature engineering; automated ETL and warehousing in Python and SQL, cutting processing time by 70% and shipping dashboards adopted across departments.

Stack: scikit-learn, Python, SQL, ETL

Projects

Cairn — An autonomous experiment runner that remembers what it tried. Karpathy's autoresearch loop hits a wall after about forty experiments: the agent starts re-proposing what it already tried. Cairn keeps a wiki as memory, so every kept experiment is a stone on the trail. code
Shadow-agent — A terminal that learns how you work and predicts your next command, entirely on your device. It learns your hands. Nothing leaves your machine. live code
Browser-LLM — A fully client-side chatbot: WebGPU acceleration, zero backend, complete on-device privacy. live code
quark — Quantum programs can be written many ways and still compute the same unitary, and checking that exactly is exponential. quark turns each circuit into a vector so equivalent ones land close and unrelated ones sit near zero. live code
SecFix — An autonomous secure-coding agent. It reads source code, finds real vulnerabilities with a language model, rewrites the affected files, then re-audits its own patch as an independent reviewer. code
Plumbline — Language models make things up. Plumbline checks an answer claim by claim against the sources it was supposed to come from and points at the exact span that backs or contradicts each one. code
DataWiz Pro — Privacy-first AutoML and business intelligence for tabular data. Load a file, and it profiles, cleans, selects features, tunes a dozen models and explains the winner, entirely in the browser. live code
PhaseGraph — A deterministic compiler and runtime for multi-agent context engineering. Describe agents, artifacts and delivery requirements as a graph; it compiles a scheduled plan under capacity, clearance and budget constraints, executes it against a real corpus, and reconciles what it predicted with what it measured. live code
Veilcap — A Tor-native, capability-controlled personal compute network. code

Games and tools

cli-legend — roguelike dungeon crawler on npm (npm i cli-legend) link
Pixel-legends — 2D retro game hub, live link
Zelda-like — a Godot action-adventure link
gamedev-simple-utils — small utilities for game dev (npm i gamedev-simple-utils) link
dev-toolkit-sampark — a CLI toolkit for shipping fast (npm i dev-toolkit-sampark) link

Papers

An Energy-Efficient Hybrid Communication Protocol for Large-Area Wireless Sensor Networks. Procedia Computer Science (Elsevier), ICECMSN, 2025. Shreyas Karthigeyan, Sampark Bhol, Akshat Jain, Karthik Yadav, Akash Sinha, Padma Priya R. DOI 10.1016/j.procs.2025.01.059
Hybrid Classical and Quantum Query Complexity for Hypothesis Verification Under Structured Priors and Bounded Phase Noise. Journal submission, with QuantumX, 2026. Under review. Sampark Bhol.

Writing

2026-04-01 — Autoresearch: Automated ML Optimization While You Sleep (1888 words)
2026-02-11 — The Structured Militarization of Advanced Computing (1263 words)
2026-02-11 — From Models to Machines: AI and Quantum's Late 2025 to Early 2026 Developments (2176 words)
2025-12-20 — Defence Gets Faster, Sharper, and Stronger x Quantum Computing (1976 words)
2025-09-01 — QUANTAI (Quantum and Tokenized AI Sidechains): opening a possibility in this limitless field of AI (843 words)
2025-08-23 — Advancing RL in Multimodal Games: Tackling Offline, Continual, and Bias Challenges (2048 words)
2025-06-27 — The Rise of Agent Communication: Google's A2A and Anthropic's MCP (1735 words)
2025-05-20 — NeuroSymbolic AI: A View (988 words)
2025-03-05 — Quantum-Enhanced NLP: The Future of Cloud AI (988 words)
2024-10-09 — CERN in October: A Month of Milestones and Marvels (1331 words)

Certifications

EP.01 · IBM Deep Learning Fundamentals, IBM (Cognitive Class), 2024 · verify
EP.02 · ISAA Saviynt Identity Security, Saviynt, 2024 · verify
EP.03 · Building Transformer-Based NLP Applications, NVIDIA, 2025 · verify
EP.04 · Oracle Cloud Infrastructure Associate, Oracle, 2025 · verify
EP.05 · Oracle Generative AI Cloud Infrastructure Professional, Oracle, 2025 · verify

Education

B.Tech, Computer Science & Engineering, VIT Vellore, 2022 – 2026, Vellore, India.

Also

AI/ML lead. IEEE Computer Society, VIT (2023 – 2025): hackathons with 200+ participants, workshops on deep learning, NLP and computer vision.
Member. Association of Energy Engineers (2023 – 2024).
Off-hours. Browser games, a Godot action-adventure, a roguelike on npm. Games are how I learn game feel.

Skills

Languages & systems: [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object]
AI/ML & generative AI: [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object]
Data science, analytics & engineering: [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object]
Cloud, MLOps & databases: [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object]
Mobile, web & games: [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object]

Contact

Open to: Backend and platform roles; agent systems. Where: Remote, or Munich. Timezone: IST, overlaps CET. Response: 24–48 hours.