AI Solutions Engineer · London
I turn messy business problems into shipped AI systems — at a fraction of the bill you’d expect.
AWS, GCP, Azure when needed. Hetzner + Coolify when smarter. Built end-to-end, by me.
Most production work sits in private repos under client / company NDA.
Open to roles
From construction procurement to wealth management — applied AI delivered in real-world contexts across two continents.
Construction
CMS Desk · Spencer AI · Blueprint
Procurement
AutoQuote · Rebate Mgmt · POD Linking
Logistics
Way2Packers · 80+ cities
Tax & Compliance
Tax2Solutions · UK + India + USA
HR Tech
PixelEMS · 7 AI agents
EdTech
Way2Class · 25+ countries
Wealth Management
Wealth1 · India's first PMS / AIF marketplace
Marketing
SocialPixelHub · 12+ verticals
Document AI
AutoQuote · 12 agents · 95+ fields
Government
TransKai · multi-language YouTube review
Accessibility
ALT-Kai · WCAG 2.2 + BenTech
Solar / Renewables
Manav Solar Solutions · 5,000+ families · 12 MW+
Every metric traced back to a shipped system, a real client, a live dashboard. No vanity numbers.
LLM API cost reduction
£800/mo → £30/mo at CMS Desk
OSS vs premium accuracy
MSc dissertation finding
Infrastructure savings
Hetzner + Coolify over AWS/GCP
Docker containers in prod
across CMS Desk workloads
Specialised AI agents
in the AutoQuote pipeline
Years applied AI
2021 — present
Monthly LinkedIn impressions
News in Pixel · organic
Articles published
Tutorialspoint · Codedamn (+ ghostwriting)
About
Less hardcore engineer — more AI generalist and delivery lead. I pick the right model, the right tool, the right shortcut, and get the thing into production. Five years of exactly that across 12 industries.
Plate I.
Nikesh Malik
London, United Kingdom
MSc
Birmingham
I’m the person you bring in when a messy business problem needs to become a working system in production — fast.
Not the algorithm purist. Not the deepest researcher in the room. I’m the operator who knows which model to pick, which tool to reach for, and how to get the thing live this week — without breaking what’s already shipped.
At CMS Desk that means Spencer AI, AutoQuote (12-agent doc pipeline), Blueprint Analysis, and the multi-model router that cut LLM spend 96%. Through PenPixel LLP, AI products shipped across logistics, tax, HR, marketing, edtech, wealth management, and energy.
I won’t sell you an AWS bill you don’t need. I know AWS, GCP, and Azure inside-out — but most of my production work runs on Hetzner VPS with Coolify, self-built, self-orchestrated, end-to-end by me. Cost-effective architecture isn’t a side-effect — it’s the choice.
“Get the smallest useful version into production. Measure it. Make it better. Repeat.”
— How I work
Cost Engineering
Three production wins that compound. The compounding effect across 40+ containers and dozens of chains is what gets you to 96%.
LLM API spend
Before
£800/mo
After
£30/mo
Routed across OpenRouter, AWS Bedrock, Azure OpenAI, GCP Vertex, and self-hosted Hetzner OSS. Simple tasks → Llama 3 / Mistral. Medium → Gemini Flash. Premium reserved for high-stakes reasoning.
Infrastructure
Before
AWS · GCP
After
Hetzner + Coolify
40+ Docker containers moved to Hetzner orchestrated by Coolify — production reliability maintained, monthly bill collapsed.
Research finding
Before
Premium models
After
Open-source · zero cost
12 LLMs benchmarked across 1,144 verified bugs and 4,320 evaluations. Open-source models achieve 92.5% of premium accuracy — the empirical justification for everything above.
Every workflow gets routed to the cheapest model that can do the job correctly. Every host runs on the cheapest infrastructure that meets the SLA. Compound that across 40+ containers — and you get to 96%.
Work
Featured case studies, products and platforms, plus notable engagements. Most production code sits in private repos under company policy.
Agentic NL-to-SQL Chatbot · Internal Platform
Natural-language chatbot that lets CMS users query their business database in plain English. Converts questions into SQL, runs them against a real-time synced copy of the production DB, and returns formatted answers with charts, exports, and follow-up suggestions.
Multi-Tenant Document AI · 12 Specialised Agents
Multi-tenant document AI platform. Ingests invoices, receipts, PODs, credit notes, quotations, subscriptions, and bank statements via email or upload, classifies them, extracts structured data, and stores results.
Cost Engineering · Production Routing
Production routing layer across 6+ providers — open-source (Llama, Mistral, DeepSeek) and commercial APIs (GPT, Claude, Gemini) — with intelligent task-complexity routing, automatic fallback, and per-call cost telemetry.
Computer Vision · Construction Procurement
Automated material take-off from 2D/3D construction blueprints. Detects rooms, doors, windows, fixtures — outputs a procurement-ready material schedule with quantities.
Accessibility AI · Multi-Step Gemini Agents
AI alt-text generator for technical and research content. Ingests PDFs, Excel sheets, or batches of images and outputs WCAG 2.2 + BenTech-compliant alt-text using a multi-step Gemini agent pipeline.
Multi-Lingual Video Intelligence · Brand & On-Screen Detection
AI agent that downloads any YouTube video, auto-detects the language, transcribes with Whisper, translates with Gemini, AND simultaneously watches the video to detect on-screen text, brand mentions, sponsor placements, and speaker turns. Outputs a multi-column spreadsheet for downstream review.
Agentic OS for UK SMEs · 7 AI Agents
Multi-tenant Agentic OS for UK mid-market operators. Consolidates 5–7 disconnected HR/compliance/ops tools into one platform with 7 specialised AI agents and human-authorised workflows.
Logistics Platform · AI Chatbot · Deep SEO
Full-stack logistics platform with AI chatbot, deep local SEO architecture, and quote workflows. 400+ pages serving 80+ Indian cities and 50+ neighborhood-level locations.
AI Marketing Automation · Multi-Model Content
AI-powered marketing automation platform combining social media management with SEO. Scans a website in 60 seconds, learns the brand, then auto-generates and posts on-brand content across 6 platforms.
AI Tax Automation · India + USA + UK
CA-backed tax, compliance, and company registration platform across India, USA, and UK. Multi-step LLM agents process bank statements, classify transactions, calculate taxes, and draft filings.
India's first AI-Driven PMS / AIF Marketplace for HNIs & NRIs
India's first AI-driven PMS (Portfolio Management Services) and AIF (Alternative Investment Fund) marketplace targeted at High-Net-Worth Individuals and Non-Resident Indians.
Rooftop Solar Platform · PM Surya Ghar Yojana · Full-Stack Build
Full-stack web platform for a rooftop solar installer in India — government subsidy calculator, site survey booking, services / products / pricing / areas / FAQ pages, integrated with the PM Surya Ghar Yojana scheme. Built end-to-end including design, development, and deployment.
.png&w=3840&q=75)
Internal Platform · Built from Scratch
Custom Prometheus + Grafana observability for every LLM workflow — cost per provider per day, P50/P95/P99 latency, error rates, fallback triggers, token usage per chain per model.
.png&w=3840&q=75)
Custom Workflow Automation Platform · Content + Social
Self-built workflow automation platform — visual content pipeline (Schedule → Brief → AI Draft → SEO → Approval → multi-channel publish). Replaces n8n / Zapier with a focused OSS-friendly stack tuned for content + social ops at CMS Desk.
Automated News Platform
Automated short-form news platform aggregating last-24hr news. Grown organically to 100K+ monthly LinkedIn impressions in 6 months.
Internal HR Tool · Multi-Company
Internal employee management platform used across multiple companies for attendance, payroll, and leave management.
Edtech Delivery Oversight · 25+ Countries
Educational solutions delivery — eLearning, AR/VR, accessibility (WCAG 2.2, ADA, EAA), instructional design, localization across K-12, higher education, corporate training.
Empirical Research · Distinction
Comprehensive empirical analysis of 12 LLMs and 3 static analysis tools across 1,144 verified bugs using a four-dimensional evaluation framework (accuracy, speed, cost, semantic understanding). 4,320 primary evaluations with full statistical validation.
Experience
Five roles, five years of learning — AI engineer, founder, ops manager, and freelance technical author across two of India's largest developer-education platforms.
AI Solutions Engineer — Delivery & Implementation
London, UK · On-site · Full-time
Designing and delivering applied AI for construction and procurement — scoping with stakeholders, architecting solutions, managing end-to-end delivery. 40+ Docker containers in production on Hetzner with custom Prometheus/Grafana observability across every chain.
Founder — Technical Delivery & Product Management
United Kingdom & India · Remote
Founded and lead a digital solutions company delivering AI-powered products, websites, and automation. End-to-end delivery from requirements through launch across logistics, tax, HR, marketing, edtech, and wealth management.
Technical Operations Manager
United States · Remote · Full-time
Managed technical operations and cross-functional delivery — coordinating frontend, backend, marketing, and SEO teams across the organisation.
Technical Content Writer · Freelance
India · Remote
Wrote tutorials, code walkthroughs, and developer-facing documentation across SQL, Python, JavaScript, React, Node.js, database management, and full-stack web development for two major developer-learning platforms.
Stack
Production-tested across five years of shipping AI systems.
AI-Assisted Coding
Every line of production code at CMS Desk and PenPixel ships with AI agents in the loop. Daily-driver tools — not just on the CV, but in the terminal.
Primary terminal-native coding agent
Anthropic's official CLI — full repo context, multi-file edits, agent SDK
Async background tasks
Cloud agent for delegated long-running work, PR review
Open-source coding agent
Self-hosted alternative for sensitive client work
Git workflow automation
PR drafts, commit messages, code review at the terminal
Conversational reasoning
Interactive thinking partner for architecture decisions
Agentic routing layer
Open-weights model used in the multi-model router fallback chain
Local LLM serving
Self-hosted Llama / Mistral / DeepSeek runtime on Hetzner
[ The point ]
Magnetic developer means knowing which agent to reach for — terminal-native for repo work, async for delegated tasks, self-hosted when the data is sensitive. The right tool for the right cut of work.
Voice
From recent LinkedIn posts — unfiltered, written from production experience at CMS Desk.
Embedding quality matters more than model size. Hybrid search catches what pure semantic search misses. Metadata filtering before vector search dramatically reduces latency. Qdrant Essentials cert: 100% score (20/20).
The agent isn't the hard part — knowing whether it's actually working is. Custom Prometheus/Grafana dashboards: cost per provider, P50/P95/P99 latency, error rates, fallback triggers, tokens per chain per model.
Structured output parsing. Tool calling architecture. Retrieval chains across pgvector + Qdrant. Multi-model routing across providers — that's how we cut LLM API spend from £800/mo to £30/mo.
Stateful multi-agent workflows. Explicit routing control. Human-in-the-loop checkpoints. Retry logic at each step. Used at CMS Desk for document classification, RAG retrieval, quote generation, banking reconciliation.
Writing
Tutorials and developer documentation published under my name on Tutorialspoint and Codedamn. A much larger volume written under freelance agency contracts and ghostwritten for other publications. Plus two peer-reviewed IRJET papers.
Dec 2022 — Jul 2024
Step-by-step tutorials covering SQL, Python, database management, web development, and software engineering fundamentals — published on one of the world's largest tutorial platforms.
2023 — 2024
Hands-on coding tutorials and developer guides on JavaScript, React, Node.js, Python, and full-stack web development for the codedamn learning platform.
2023 · 2024
Two peer-reviewed research papers in the International Research Journal of Engineering and Technology (Impact Factor 8.226). BERT-based cybersecurity detection and multimodal social-media NLP.
Education
MSc Birmingham (Merit), B.Tech Mumbai (First Class CGPI 9.01), four professional certifications from LangChain Academy and Qdrant, plus two peer-reviewed IRJET papers.
Education
MSc Advanced Computer Science
2:1 Merit · Distinction in Dissertation (75/100)
Dissertation: "Open vs Closed LLMs vs Traditional Linters: A Comprehensive Empirical Analysis Across Four Critical Dimensions" — supervised by Prof. Rami Bahsoon. Benchmarked 12 LLMs and 3 static analysis tools across 1,144 verified bugs using a novel four-dimensional framework. 4,320 primary evaluations with full statistical validation (ANOVA, Welch's t-test, Cohen's d). Key finding: open-source LLMs achieve 92.5% of premium model accuracy at zero cost.
Selected modules
B.Tech Computer Engineering
CGPI 9.01/10 · UK First Class · US GPA 3.6/4.0
Specialisation in Blockchain & Cybersecurity. Smart India Hackathon participant 2022–2024. 2nd place, Inter-College Business & Startup Competition. Final semester SGPI 9.14/10 (Project: Grade O, Industry Internship: Grade O).
Certifications
Qdrant Essentials
Qdrant Academy · 100% Score (20/20)
ID: QDRANT-1E3F62CC
Agent Observability & Evaluations
LangChain Academy · LangSmith tracing, eval metrics, debugging production agents
ID: zbwgw7xzv5
LangChain (Python)
LangChain Academy · Production patterns
LangGraph
LangChain Academy · Stateful multi-agent workflows
Working philosophy
Certificates prove curiosity. Shipped systems prove competence. I aim to do both — keep learning, keep delivering.
Open to permanent roles, contract engagements, and advisory work in applied AI, AI engineering, and technical delivery. Based in London, available immediately.
— Nikesh.
