Services
AI engineering for teams that need it to work in production.
I work with founders, product teams and engineering leads who have a real workflow to automate or a knowledge problem to solve — and who want a system that survives contact with real users, real data and real costs.
AI Agents & Workflow Automation
Multi-step agents that execute real business processes — outreach, scheduling, research, coordination — with explicit state, tools and human hand-off.
Problems solved
- Teams spend hours on repetitive multi-step workflows across email, calendars, CRMs and internal tools.
- Existing automations break as soon as a step needs judgement.
- Long-running processes have no memory, recovery or audit trail.
What you get
- ↳LangGraph / CrewAI agent architecture
- ↳Tool and API integrations
- ↳State persistence, retries and human-in-the-loop controls
- ↳Containerised deployment
RAG & Enterprise Knowledge Assistants
Grounded assistants over your documents, regulations, policies and internal knowledge — with citations, access control and a refresh pipeline.
Problems solved
- Critical knowledge is buried in long documents nobody has time to read.
- Generic chatbots hallucinate and cannot cite sources.
- Knowledge bases go stale and nobody trusts the answers.
What you get
- ↳Ingestion and chunking strategy
- ↳Embedding and vector-store design
- ↳Retrieval + generation pipeline with citations
- ↳Evaluation harness and refresh workflow
Custom LLM Applications
Product features and internal tools powered by LLMs — personalisation, summarisation, classification, conversational interfaces — engineered for consistency and cost.
Problems solved
- You know an LLM could add value but the prototype is not reliable enough to ship.
- Outputs are inconsistent, expensive or off-brand.
- The feature needs memory, structure and guardrails.
What you get
- ↳Prompt and context engineering
- ↳Structured outputs and guardrails
- ↳Session memory and personalisation
- ↳API layer and monitoring
Machine Learning Solutions
Forecasting, regression, anomaly detection and classification pipelines built on your data, with honest validation and simple interfaces.
Problems solved
- Forecasts and anomaly checks are done manually or not at all.
- Every ML request depends on a data scientist being available.
- Models exist but nobody trusts or uses them.
What you get
- ↳Data profiling and feature pipeline
- ↳Model training, validation and selection
- ↳Automated retraining workflows
- ↳Plain-language reporting
Computer Vision & OCR
Detection and document-understanding pipelines — labels, forms, parcels, invoices — that turn images into structured data.
Problems solved
- Operational data is trapped in photos, scans and labels.
- Manual data entry is slow and error-prone.
- Off-the-shelf OCR fails on your document types.
What you get
- ↳Object detection (YOLO) and OCR (PaddleOCR) pipelines
- ↳Post-processing and entity mapping
- ↳Accuracy evaluation on your data
- ↳API deployment
AI Backend & Production Deployment
The engineering that makes AI features dependable: APIs, containers, orchestration, monitoring and cost control.
Problems solved
- A notebook works but there is no path to production.
- AI features are slow, expensive or fail silently.
- Nobody knows how the model is performing in the wild.
What you get
- ↳FastAPI service design
- ↳Docker and Kubernetes deployment
- ↳Observability with Grafana
- ↳Latency and cost optimisation
How an engagement works
Clear scope, working software every week, clean hand-over.
01
Discovery
A short call to understand the workflow, data, constraints and what success looks like in numbers.
02
Architecture & plan
A written proposal: system design, model and retrieval strategy, evaluation approach, milestones and cost drivers.
03
Build in the open
Weekly demos of working software, shared repo, and honest evaluation results — not slide decks.
04
Ship & hand over
Containerised deployment, monitoring, documentation and a walkthrough so your team owns it after I leave.
Availability
Open to remote AI/ML roles and select consulting engagements
IST (UTC+5:30) · overlaps US, EU & UAE hours