Arpan Ghoshal
Short versionEverything
CVBook a call

Everything, in one place.

Every role, skill, product and paper. The full CV is a click away.

What you'd be hiring

Applied AI and Forward Deployed Engineer with 5+ years building production AI systems: AI agents, LLM applications, RAG, evals and guardrails, NLP and machine learning. The short version is below; the full story is on my CV.

Director of AI Innovation

For Halsey Minor's ventures. I led a team of 12, built two 0-to-1 AI products end to end, and demoed them to secure continued investor funding.

ctrlrun

My open-source execution-safety layer for AI agents, and an official LangChain middleware integration.

96% accuracy

Gisual's first commercial machine learning model for power outage detection, deployed for utility clients.

3 startups upgraded

Production LLM and RAG systems in Europe and the US: lower inference costs, better accuracy, and eval suites to prove it.

OWASP GenAI Security Project

Section author on non-repudiation and audit trails for AI agents, in the Agentic Security Initiative.

Ready to start

US (approved H-1B, transfer), UK (HPI) and the Netherlands (Zoekjaar), with no sponsorship needed in the UK or the Netherlands. Available now.

Where I've worked

  1. Applied AI Engineer (Independent Consultant)Mar 2026 – Sep 2026
    Independent AI Engineering & Consulting

    Upgraded production AI systems for 3 startups across Europe and the US, from scoping to handover.

  2. Director of AI InnovationFeb 2025 – Feb 2026, retained to Sep 2026
    Operating Equity Partners (Vivid Labs Holdings, Inc.)

    Led a team of 12 with Halsey Minor and built two 0-to-1 AI products end to end.

  3. Machine Learning EngineerSep 2023 – Feb 2025
    Gisual Inc.

    Built the first commercial machine learning model for power outage detection, at 96% accuracy.

  4. Machine Learning EngineerSep 2021 – Jun 2022
    Hitch Works (acquired by ServiceNow)

    Built a BERT-based skills matching model, 170% more accurate than the baseline.

  5. Machine Learning EngineerJan 2021 – Aug 2021
    Brainentech Neuroscience

    Built and deployed a real-time audio analysis API serving 150k+ users.

  6. Data Engineer InternJun 2020 – Aug 2020
    NayaOne

    Built the company's first data infrastructure and data APIs, later adopted by major fintech clients.

What I work with

Agents & LLMs
Claude Code (power user), MCP servers and gateways, LangChain, LangGraph, OpenAI Agents SDK, OpenClaw, OpenAI/Anthropic APIs, tool calling, structured outputs (Pydantic), RAG, Pinecone, ChromaDB
Evals & Safety
Evals, guardrails, human approval gates, agent audit trails, LangSmith, Langfuse, OpenTelemetry
ML
PyTorch, Hugging Face Transformers, fine-tuning (LoRA/QLoRA), scikit-learn, XGBoost, TensorFlow
Languages & Backend
Python, TypeScript, FastAPI, Flask, REST APIs, WebSockets
Cloud & Data
AWS (SageMaker, Lambda, S3, EC2, ECS/EKS, RDS, Step Functions), Cloudflare, Docker, Kubernetes, Terraform, GitHub Actions, PostgreSQL, Redis, MongoDB, Elasticsearch, MLflow

Education

  • New York University2022 – 2023

    MS in Computing, Entrepreneurship and Innovation, Courant and Stern. Completed the two-year program in one year.

  • PES University2017 – 2021

    BTech in Computer Science and Engineering, Data Science specialization.

Writing and recognition

  • Non-repudiation, Immutability, and Retention

    OWASP GenAI Security Project, Agentic Security Initiative, 2026

  • An Enhanced Context-Based Emotion Detection Model Using RoBERTa

    IEEE CONECCT, 2022

  • CATAQ: Concise Answer to Any Question

    ICSES, 2021

  • AWS Certified Solutions Architect
  • NASA Space Apps Challenge Mentor & Judge
Read my CVLinkedInGitHubHugging Face

What I'd do for your product

Take it from 0 to 1

From a rough idea to a working AI product: scoping, architecture, agents and retrieval, evals, deployment and handover. I work directly with founders.

Fix what isn't working

If your LLM or RAG system is slow, expensive or inaccurate, I audit it, re-architect what needs it, build an eval suite, and ship the improvements.

Evals you can trust

Golden datasets, LLM-as-judge scoring, regression tests, guardrails and tracing of quality, latency and cost per request, so you know whether a change made things better before it ships.

Agents and automation

AI agents, MCP servers and workflow automation on top of your stack, with human approval on the actions that matter.

Results

  • Two 0-to-1 AI products for Halsey Minor's ventures, which secured continued investor funding.
  • Production AI systems upgraded for 3 startups in Europe and the US.
  • ContractReviewer, which PandaDoc's CEO and CTO proposed integrating into their platform.

Founders I've worked with: fintech, HR tech, utilities, neuroscience, legal and news.

How we'd start

  1. 1
    Book a call
    Tell me what you're building. 15 minutes to an hour, your pick.
  2. 2
    Scope it together
    Agree on the problem, what done looks like, and how to measure it.
  3. 3
    Build and ship
    I build, evaluate and hand over, working with your team the whole way.
Book a call

Things I've built

ctrlrun

Execution safety for AI agents. A person approves the exact action, each effect runs at most once, and every action leaves a receipt.

ctrlrun.devGitHub

ctrl ai agents

A control room for every AI agent a company runs. See each action, and allow, ask or block it before it runs.

ctrlaiagents.com

ctrl payments

The same boundary for money. Refunds and payouts an agent attempts wait for a person to approve the exact amount.

ctrlpayments.com

iwrote

Certifies writing as human by recording how the words arrive, instead of an AI detector's guess.

iwrote.co

Puchku

A privacy-first wellbeing companion with on-device AI and end-to-end encrypted sync. Launching soon.

puchku.app

ContractReviewer

Flags contract risks, hidden fees and one-sided clauses in plain English.

contractreviewer.io

What I believe

  • Build what people actually use.
  • AI assists, people decide.
  • Prove what's human, protect what's private.