MLOps Engineer · Coventry Building Society · Coventry, UK
I take machine learning models out of notebooks and into production — versioned, governed and monitored, on platforms a regulated lender is willing to bet on. Credit risk models, agentic AI, and the pipelines that keep both of them honest.
# the short version
apiVersion: mlops/v1
kind: Engineer
metadata:
name: jacob-binu-mancherikalam
location: coventry, uk
spec:
role: MLOps Engineer
domains: [credit-risk, agentic-ai, llmops]
clouds: [aws, azure, gcp, databricks]
runtime: [python, pyspark, docker, kubernetes]
status:
phase: Running
since: 2025-01stage · profile
Most machine learning work dies somewhere between a data scientist's experiment and a system people actually use. That gap is the job I do. I build the end-to-end lifecycle around a model: versioning and reproducibility, CI/CD across dev and prod, serving, monitoring, and the governance a regulated environment demands.
At Coventry Building Society I deploy IRB and IFRS9 credit risk models into the analytics platform, engineer serverless ETL into Redshift with AWS Glue, and lead Proof-of-Value work on enterprise AI platforms — with explainability and responsible AI as first-class requirements, not an afterthought.
Outside the day job I build the same thing at small scale, in public: agent frameworks with LangGraph, GraphRAG systems, Databricks MLOps pipelines. The repos below are the receipts.
stage · experience
Coventry Building Society · Coventry, UK
Tata Motors Design Tech Centre · Coventry, UK
Mancherikalam Supermarket · Kottayam, India
stage · projects
Every one of these is public. Clone them, break them, read the pipeline code.
A cloud-native AI chat interface powered by Google Gemini — an interactive workspace for evaluating agentic workflows with causal reasoning, live code execution and web search, behind a production-grade GCP deployment pipeline.
View repository ↗An agent framework built from the ground up with LangGraph and Claude: task planning, virtual file systems and sub-agents for multistep reasoning. ReAct loops, Pydantic state management, Tavily search, async Python orchestration.
View repository ↗A GraphRAG system in Flask, wiring the Ollama API to local models so retrieval reasons over graph structure instead of flat chunks — better context, fewer confidently wrong answers.
View repository ↗A minimal, production-ready ML workflow on Databricks: data prep, training and tracking with MLflow, model registration, and batch inference and serving. The core MLOps stages, with nothing hand-waved.
View repository ↗Enterprise-grade MLOps on Databricks for real-time anomaly detection in finance: streaming ingestion, automated retraining and continuous deployment — a model that stays current instead of quietly rotting.
View repository ↗A scalable MLOps system classifying plant disease (healthy, powdery, rust) with CNNs — research code turned into an automated pipeline with a real deployment and lifecycle story.
View repository ↗A Flask app that reads sentiment from top financial voices in real time, tracking market mood and surfacing trends on a live dashboard.
View repository ↗stage · skills
stage · education
Coventry University · 2021 — 2024
Dissertation — Contextual Intelligence in LLMs via a RAG Framework. Built an end-to-end retrieval-augmented generation pipeline over OpenAI and Meta models with a vector database, tuned prompting and chunking strategies for retrieval accuracy, and benchmarked proprietary GPT against open-source Llama on cost, latency and quality to inform model selection.
Artificial Neural Networks & Intelligent Agents · Machine Learning & Applications · Advanced Algorithms · Big Data Programming · Security · Theory of Computation · Software Engineering
stage · contact
Hiring for MLOps, platform or applied AI work? I reply to everything.