Shoten Deshmukh

Applied AI & Solutions Architecture

I help startups and enterprises get AI into production. I have built agentic systems under regulation, founded a startup, and deployed Claude into external organisations as a trusted technical advisor.

shotend@gmail.com+44 7792 338583Londongithub.com/shoten-healthjourneys

Capabilities Agent

You could read my CV. Or you could ask an agent instead.

Claude Sonnet 4.6 backed by 8 MCP tools, retrieving structured evidence in real time. Built with Claude Code. This site is itself a demonstration of the agentic patterns described below.

Capabilities Agent
Hi, I'm an AI agent built to answer questions about Shoten's experience and fit for the Solutions Architect, Applied AI (Startups) role. I use tools to retrieve structured evidence. Ask me anything, or try one of the suggestions below.

What I bring

Six pillars, each backed by production evidence

Pillar

“Builder credibility: technical founder who ships”

Evidence

Founded Health Journeys and built the entire production platform solo with AI tools. Won partnerships through founder-led sales at CEO level. Built gym-trainer-ai (Claude Sonnet, 10 MCP tools, full production stack). Four production agentic AI solutions at Benchmark, all hands-on architected.

Pillar

“Trusted technical advisor to founders and engineers”

Evidence

Own customer relationships across 150 adviser firms at Benchmark: lead technical discovery sessions, run demos, and act as the senior technical voice throughout the adoption journey. Act as an unpaid solutions architect for startups on evenings and weekends, deploying Claude Code and advising on AI adoption.

Pillar

“Hands-on with LLMs in production: context engineering, evals, modern architectures”

Evidence

Designed the context engineering pipeline for the Fact Find Agent: document ingestion, CRM retrieval via tool use, ontology grounding, HITL clarification. Built evaluation frameworks for every AI initiative at Benchmark. Built gym-trainer-ai with Claude API and 10 custom MCP tools.

Pillar

“Wins technical evaluations and sells to technical buyers”

Evidence

Actively involved in the sales process at Benchmark: pitch the platform to prospective firms, lead technical demos, shape proposals that win new business. At Health Journeys, ran founder-led sales to close four hospital partnerships at CEO level.

Pillar

“Gathers insights and feeds back to Product and Engineering”

Evidence

Gather deployment insights across 150 firms, identify emerging patterns, and feed technical requirements back to product and engineering to shape the roadmap. Wrote reusable technical playbooks so every new onboarding was faster than the last.

Pillar

“Passionate about safe, beneficial AI”

Evidence

Every agentic solution has human escalation and compliance-in-the-loop built in from the start, not bolted on. 7+ years navigating FCA regulation, security reviews, data residency. You measure, govern, then ship.

What I've Built

Health Journeys
Problem

UK patients stuck on NHS waiting lists. World-class healthcare in India at a fraction of the cost. No trusted bridge.

Approach

Founded and run the company alongside my Benchmark role. Built the entire production platform solo using AI tools: 30+ page website, AI-powered patient chatbot on WhatsApp and Messenger, automated blog publishing, analytics, lead capture, and paid acquisition. Built a multi-agent marketing team that automates content production: a manager agent coordinates a content strategist, content creator, and reviewer agents to produce and publish marketing content autonomously. Negotiated CEO-level partnerships with Gleneagles Healthcare (6 locations, 1,200+ beds) and Zydus Hospitals (1,100+ beds, internationally accredited).

Result

4 hospital partnerships, active patient pipeline, automated marketing engine, full platform built and operated by one person with AI. Proof that one person with the right tools can ship what used to take a team.

Patient
Website (30+ pages)
AI Chatbot (WhatsApp / Messenger)
Lead Capture
Hospital Partners
Fact Find Agent
Problem

Advisers spend ~90 mins per review on manual admin across 34,000+ annual reviews

Approach

Agentic AI that analyses meeting notes and financial PDFs, retrieves CRM context via tool use, grounds against a financial planning ontology, runs HITL clarification

Result

~90 min reduction per review. Production under FCA regulation.

Meeting Notes + PDFs
Agent
CRM via Tools
Ontology
HITL Check
Adviser Review
CRM Updated
Advice Case Orchestrator
Problem

Advice cases need complex workflow management with compliance checkpoints

Approach

Multi-agent system with tool use for task management, compliance review agents, human escalation

Result

Compliance-by-design in the critical path. Production deployment.

Adviser
Orchestrator Agent
Task Management Tools
Compliance Agent
Pass / Fail Routing
Human Reviewer
Meeting Summary Agent
Problem

Advisers spend 30-45 minutes after each client meeting writing up notes, action items, and follow-ups in a compliance-ready format

Approach

Deploys a meeting bot to capture transcripts, then summarises into a compliance-ready format with action items and follow-ups fed directly into the CRM

Result

Saves 30-45 minutes per client meeting. Production under FCA regulation.

Client Reviews Agent
Problem

Annual client reviews are complex, multi-step processes requiring preparation, meeting management, and report generation

Approach

Supports the annual client review process end-to-end: prepares the advice team ahead of the meeting, manages workflow and tasks through the review cycle, and generates the suitability report

Result

Full end-to-end support for the annual review cycle. Production under FCA regulation.

AI-Augmented PDLC
Problem

Traditional product development cycles too slow for AI-era delivery

Approach

Spec-driven development lifecycle with AI augmentation at each stage

Result

4x cycle time improvement. Now used across Benchmark’s engineering org.

Claude in the Field

I already do forward-deployed AI enablement. Voluntarily. On weekends.

Covecta.io

Set up repos, context files, Claude Code workflows. Hands-on training for the engineering team. Ongoing support.

Nikura Essential Oils

Trained CEO and leadership on Claude Code. CEO now saves 15 hours per week.

Sigma Sports

Set up Claude Code tooling and workflows for the dev team.

Same pattern each time: train leaders first, set up repos and context files, train the team, then ongoing weekend support.

Open Source

gym-trainer-ai

github.com/shoten-healthjourneys/gym-trainer-ai

Full-stack AI personal trainer. Claude Sonnet 4.6 via Microsoft Agent Framework, 10 MCP tools, Deepgram voice, React Native, FastAPI, Azure PostgreSQL, Terraform, GitHub Actions CI/CD. Built end-to-end with Claude Code.

Claude Sonnet 4.6React NativeFastAPIAzureTerraformMCP

flight-hacker

github.com/shoten-healthjourneys/flight-hacker

Parallel Docker-based flight price scanner. 8+ containers through different VPN locations to detect geo-pricing on Google Flights. Playwright, SQLite, web dashboard.

DockerPlaywrightSQLiteVPNPython