Skip to content
Sohaib Sohail
Open to work · senior, staff or principal · UK remote or hybrid

I take AI agents from prototype to production, and build the platform they run on.

Salam, and hi, I'm Sohaib. I've spent the last few years doing platform engineering at some big firms, and I've reached the point where I'd rather build my own things than only ship other people's. Most people consume what gets built. I want to be on the other side of that, making things that work and showing other engineers how I got there.

projects hosted on the AI platform I helped build
200+
clouds in production, Terraform for each
3
Google Award, GenAI Excellence
2024
01

Ask the agent

The trace below is scripted, not live. The real agent is deployed and free to try; the public instance replays recorded runs, so it costs nothing.

portfolio-agent Scripted walkthrough Open the live agent →
Try one
YOU

Open the live agent →

02

Selected work

Two flagships, deployed and open to anyone, with the depth behind them. Then the rest.

sre-investigation-agent LIVE

Incident agent with five SRE tools, streaming each call as it investigates.

  • Python
  • Bedrock AgentCore
  • Terraform
  • SSE
  • 8/8 eval scenarios passing, scored on tool choice and on the facts reaching the answer
  • One scenario has a false premise, to catch an agent inventing a cause rather than tracing the real dependency
  • Deterministic mock infrastructure so eval scoring stays reproducible
  • Public demo replays recorded runs, so anyone can try it at zero Bedrock cost

mcp-context-inspector LIVE

An MCP server that makes context window usage visible for any agent loop.

  • Python
  • MCP
  • DynamoDB
  • Firestore
  • OTLP
  • 300 automated tests, run on every push, no live cloud credentials needed
  • Three storage backends behind one data-access layer: SQLite, DynamoDB, Firestore
  • Ingests OTLP telemetry from Claude Code and Copilot, with a precision-biased redaction pass
  • Full OAuth discovery (RFC 9728) plus Google sign-in, so any MCP client can connect

gcp-vertexai-app

Vertex AI Vector Search indexes and endpoints, with Python for embeddings and validation.

  • Terraform
  • Python
  • GCP
source ↗ 2026
03

Case studies

Platform work is mostly invisible, so here it is written up the way I would write a postmortem.

Problem
Legitimate user queries were getting blocked in production. Everyone assumed prompt-attack detection was firing on benign input, and acting on that would have burned the week.
What I did
Built 27 test cases and 45 positive controls, ran them per region, then repeated them on identical inputs to see whether the detector was even deterministic.
Result
It was the HATE filter strength, not attack detection. PROMPT_ATTACK turned out to be non-deterministic even at LOW (83% one run, 94% the next), so it cannot gate a release. One config change across three regions fixed it.
Problem
Deployments were manual, per model and per region, and you checked quota by asking someone. If it did not fit you found out halfway through, with the estate left in a partial state.
What I did
A five stage Python pipeline (quota discovery, capacity weighting, dry run, tfvars generation, apply) with one YAML file per model family driving Terraform.
Result
50+ deployments from declarative config and 30 minutes down to 2. More moving parts than a hand written resource block, which is the trade, but failures moved to the start where they are cheap.
Problem
Agent demos are usually unfalsifiable. If the data shifts run to run you cannot score it, and if the reasoning is hidden you cannot tell competence from luck.
What I did
Deterministic mock infrastructure so eval scoring stays reproducible, five tools (services, metrics, logs, deployments, cost), and a UI that streams each tool call live instead of showing a spinner.
Result
8/8 eval scenarios passing. One has a false premise: ask which deployment broke checkout-api when none did, and a good answer traces it to the inventory-service dependency rather than inventing one.
Problem
Most agent observability re-shows you data your own interface already had. The genuinely invisible part is what entered context, and how much of it the user never sees.
What I did
A drop in MCP server over Streamable HTTP with eight tools, so any client can connect, Claude Desktop included. Token counts are labelled estimates rather than exact usage, documented as a deliberate trade.
Result
System prompt, tool specs, reasoning, results and answer, in the order they actually entered context, split by what the user sees versus overhead.
04

Experience

Four roles, all delivery-facing. The through-line is platform work other engineers depend on.

2026 Veracross Senior AI Platform Engineer. Run three agentic AI agents on AWS Bedrock and Ruby on Rails, turning natural-language questions into governed SQL and dashboards. Built the guardrail eval suite (64 adversarial prompts across the OWASP LLM Top 10), an LLM-as-judge harness gating every release, and cross-region failover with per-tenant token budgets. 64-prompt guardrail suite, OWASP LLM Top 10
2025–26 LSEG Senior Cloud & AI Engineer. Helped design and roll out LSEG’s enterprise AI platform on Azure AI Foundry, hosting LLMs and agentic workflows for 200+ projects. Automated deployment of 50+ LLMs, cutting rollout from 30 minutes to 2, and built an MCP server for a US team’s agentic workflow. Escalated weekly to Microsoft’s AI Product team and demoed to the CTO. Demoed live to LSEG’s CTO
2023–25 PwC Senior DevOps Engineer. Shipped an AI compliance app into NatWest, Barclays and Deutsche Bank through five client POCs to production: document comparison, summarisation and RAG based Q&A over 10+ Azure OpenAI deployments. Built the Python backend on Cosmos DB and Azure AI Search, and wired APIM and Application Insights observability across 15+ Azure AI services.
2021–22 Capgemini DevOps Engineer. Serverless and EKS workloads on AWS Lambda and API Gateway, cutting operational cost by 25%. CI/CD across Azure DevOps, Jenkins and Drone CI, with DevSecOps scanning from Anchore and SonarQube.
05

Recommendations

From people who managed me or shipped alongside me.

Sohaib, you are a delight to work with and you have continuously been supporting and helping us onboard onto this LLM platform. We’ve been able to automate our workflows and increase productivity by 45%.
An engineer who onboarded onto the platform, 2026
I had the pleasure of managing Sohaib directly, and he consistently impressed me with his strong technical capability, curiosity, and willingness to take on complex challenges. Sohaib is particularly strong in DevOps, automation, AI, and modern engineering practices. One of his biggest strengths is how quickly he learns new technologies and turns that knowledge into practical solutions. He contributed to several innovation initiatives within the team, bringing new ideas, experimenting with emerging technologies, and helping us improve the way we designed and delivered solutions. Beyond his technical skills, Sohaib is proactive, dependable, and always keen to collaborate and share knowledge with others. I would highly recommend him to any organisation looking for a strong engineer who combines solid DevOps expertise with a genuine passion for AI, automation, and continuous learning.
Anirban Chakraborty AI Architecture & Engineering, PwC UK Managed Sohaib directly
I had the pleasure of working alongside Sohaib on the AI-powered regulatory compliance app at PwC. His technical leadership was central to transforming complex compliance requirements into a production-ready application. Sohaib architected the core RAG platform for document comparison, summarisation, and policy mapping. He built the Python backend, integrated vector storage and semantic search, and automated document ingestion pipelines, slashing indexing time by 50%. What sets Sohaib apart is his dual expertise in AI and cloud engineering: he authored modular Terraform code, built CI/CD pipelines, and implemented API Management with Application Insights for full MLOps observability into latency, token usage, and model performance. He consistently delivered secure, high-performance infrastructure and solved tough technical constraints with ease. Any team would be lucky to have Sohaib, and I’d welcome the chance to work with him again!
Sharath Chandra Cloud Data Architect Worked with Sohaib on the same team
06

Work with me

I take a limited number of 1:1 sessions outside of work, for people who want a steer from someone a few steps further down the same road.

Career guidance

Breaking into platform or AI engineering, levelling up to senior, or working out your next move.

Mentoring

Ongoing 1:1 support on the engineering, the projects, and the bits nobody teaches you.

Tech consulting

Agents, MCP, LLM evaluation, guardrails, Terraform at estate scale. The things I do at work.

Book a session →
08

Background

I’m a senior cloud and AI platform engineer based in Preston, in the UK. Five and a half years in, most of it inside regulated environments where the work has to actually hold up.

Outside the terminal I follow tech news and world affairs closely, and I travel as much as I can, more than 25 countries so far. New places, new food, new context for how other people build and live.

When I’m not working I’m usually learning a new skill, building something of my own, or talking to people about what they’re working on and where it might go.

Sohaib Sohail
On a bamboo bridge in Ubud, Bali
Ubud, Bali · one of 25+ countries so far
Tech I reach for
Clouds AWSAzureGCP
Infrastructure TerraformKubernetesDockerHelm
AI & agents BedrockAzure OpenAIMCPVertex AIGuardrailsLLM eval
Languages PythonRuby on RailsSQLJavaScriptBash
Platform GitHub ActionsDatadogCloud RunECSEKSAPIM
Certifications
Terraform AssociateAzure FundamentalsAWS Certified Cloud Practitioner
Education

BSc Communications and Media Studies
Brunel University of London