Projects · production systems I designed and shipped

Projects with numbers attached.

AI automation, observability, cost control and data platforms. Each project lists what I built and what changed. Client and employer names are left out to protect confidentiality.

Filter by type

2026

Voice Intelligence Platform

Pipeline that takes customer feedback calls, transcribes them, translates multilingual speech into English and turns each call into structured insights with LLMs. A live dashboard shows sentiment, themes and top issues.

500+ calls analysed per day
Full coverage instead of manual call sampling

LLMsSpeech-to-textTranslationEvent-drivenAnalytics dashboard

2024 to now

Autonomous PR Review and CI/CD Root-Cause Analysis

A 4-agent system (Triage, Code Review, Root-Cause Analysis, Verification) on Claude API, LangChain, CrewAI and Google ADK, orchestrated through SQS/SNS with Redis state. It flags architectural flaws and security risks before a human reviews the PR, and explains failing builds.

PR review turnaround down ~35%
~25% of failing builds auto-resolved
Mean debug time down 60% with custom GitHub Actions

Claude APILangChainCrewAIGoogle ADKSQS/SNSRedis

2024 to now

Real-Time Cloud Observability and Root-Cause Analysis

A 5-agent system (Topology, Metrics Watcher, Log Analysis, Root-Cause Correlation, Remediation) that streams metrics and logs through Kinesis and CloudWatch, keeps a live service-dependency graph in Redis, and finds the cause of incidents.

Mean time to identify: ~25 min down to ~3 min
Anomaly to root cause in under 10 seconds for 90% of incidents
~30% of recurring low-severity incidents auto-remediated
~50% fewer false-positive alerts

Claude APIKinesisCloudWatchRedisMulti-agent

2024 to now

LLM Observability and Cost Control

Production evaluation and monitoring for LLM features. A CloudWatch and Grafana dashboard tracks token cost, latency and hallucination rate, so quality and spend are visible per feature.

Token spend down ~35%
Hallucination rate down 35%

CloudWatchGrafanaLLM evaluationToken cost

2024 to now

Multi-Agent Backend Task Automation

Asynchronous multi-agent systems that take over routine backend work through event-driven LLM task orchestration, with retries and isolated failures.

500+ backend tasks automated per day

LangChainCrewAIGoogle ADKQueues

Financial services client · 2021 to 2024

Bank Customer Onboarding Platform

Onboarding, Communication, Backoffice and Mutual Funds microservices in Go and FastAPI behind a React frontend, plus real-time and batch ETL on SNS, SQS, Lambda, EventBridge and Glue.

1M+ users served at 99.9% uptime
Service latency down 25%
2M+ events processed per day

GoFastAPIAWSMicroservicesETL

Consulting client · 2020 to 2021

Healthcare Big Data ETL

AWS pipeline for a healthcare client that centralises patient records using Appflow, S3, BigQuery, Glue, PySpark, Redshift and Athena, with alerts through SQS, SES and SNS.

10M+ patient records centralised

PySparkGlueRedshiftAthenaBigQuery

Consulting client · 2020 to 2021

Financial Reporting Automation

Automation for MIS, balance sheet, cash-flow and P&L reporting built on Django, Celery, Redis, PostgreSQL and PySpark, plus a Kafka and Kinesis pipeline into Redshift for large-scale analytics.

200+ hours per month of manual reporting removed
1TB+ of financial data processed

DjangoCeleryPySparkKafkaKinesis

Consulting client · 2020 to 2021

Multi-Engine AI Platform

Platform that runs several AI engines side by side, with a React frontend, payment gateway integration, user access management and customer support.

10+ AI engines running concurrently

ReactPythonPaymentsAccess control

Upwork portfolio case study

AI Customer Service Agent with n8n

Customer service agent built as an n8n workflow with Airtable and the Google Calendar API. The full case study is on my Upwork portfolio.

n8nAirtableGoogle Calendar API

See it on Upwork

Upwork portfolio

SACRD Business Intelligence and Financial Reporting Tool

Business intelligence and financial reporting automation tool. Details are on my Upwork portfolio.

ReportingAutomationBI

See it on Upwork

Upwork client project · 2025 · rated 5.0

Audio Transcription Pipeline Ported to Go

Migrated a Python-based audio processing and transcription pipeline to Go, integrated with MongoDB.

GoPythonMongoDBTranscription

Want an audit or automation like these?

Send a short email about your AI system and which service you need.

hello@aiauditor.dev

Book a 30-minute call

Back to services · LinkedIn · Upwork