Byron Mahinya Isaac

• Support enterprise customers building production workloads on Amazon Bedrock — advising on model selection (Claude, Titan, Llama), prompt engineering, RAG pipelines, agentic workflows, and cost/latency optimization.
• Triaged and resolved 307+ enterprise customer incidents across Amazon Bedrock, SageMaker, and Big Data services on AWS, maintaining a 4.8/5 customer satisfaction rating.
• Built and integrated an LLM-powered chat summary tool (Claude Sonnet 3.7 on Amazon Bedrock) into the global support dashboard, eliminating manual summary drafting for thousands of support engineers worldwide.
• Led implementation of a Feature Prioritization Model within the support tools team, establishing evaluation criteria based on security and customer impact — accelerating delivery of high-priority improvements.

• Developed and maintained the backend for an event-sourced payment and sales system processing 60+ tonnes of cooking oil monthly across 600 machines, 170 shops, and 3 refilling trucks — directly supporting the company's mission to remove 28 tonnes of plastic waste from the supply chain.
• Migrated 10+ microservices from legacy PHP to Golang, achieving a 200% increase in event throughput (180 → 430 events/min), 75% less memory usage, and 62.5% reduction in server costs ($320 → $120/day).
• Implemented distributed tracing across all microservices using New Relic, reducing system failure resolution time by 60%.
• Automated infrastructure deployment with Terraform, improving deployment efficiency by 40% and enabling rapid environment provisioning.
• Designed and implemented multi-tenant PostgreSQL architecture, improving application performance by 50% and simplifying data management.
• Owned end-to-end delivery: database design, API development, authentication/authorization, security, and production hosting.

• Built a computer vision grading module using PyTorch for counting and grading agricultural stems, improving grading accuracy by 75% over manual processes.
• Implemented gRPC client streaming for image delivery, reducing mobile data usage by 45% — critical for users in low-bandwidth environments.
• Designed and validated AI solutions aligned with business objectives, contributing to a 20% increase in revenue from AI-driven recommendations.
• Built API endpoints for interfacing with the grading module in production.

• Developed a credit scoring model using SparkML and Kafka for a Digital Sacco client, reducing loan default rates by 29% within 4 months — directly improving financial outcomes for underserved communities.
• Built Spark cron jobs transferring 10GB of data daily from cloud to on-premise servers, enabling timely access for BI and data science teams.

Built a prediction market platform for Kenya end to end: a Go automated market maker (LMSR) that prices YES/NO outcomes as live probabilities, a wallet with MPesa Daraja STK Push deposits and B2C withdrawals, real-time price updates over WebSockets, grouped events, crowdsourced market proposals, and an admin resolution flow that pays out winning positions automatically. Frontend is a server-rendered Nuxt app on Cloudflare Pages; the API runs on EC2 behind nginx.

Built a “buy me chai” tipping platform for Kenyan creators, artists and performers. Each creator gets a permanent QR code and tip page; the Go backend handles PASETO session auth, email verification, password reset, phone verification, and a secure payout-phone change flow that verifies against the old number before switching. Transactional email is delivered through asynchronous Redis-backed workers (asynq). “Erokamano” is Dholuo for “thank you.”

Designed and built a multi-tenant logistics backend serving three independent courier businesses in Nairobi, covering parcel intake, dispatch, driver assignment, and same-day delivery tracking. Ongoing support and maintenance.

SaaS revision platform for Kenya's Competency-Based Curriculum (Grade 4–12, Junior School / KJSEA emphasis). Question bank structured by the official CBC hierarchy — Grade → Learning Area → Strand → Sub-strand — with free-text answers marked against the four CBC rubric levels (EE, ME, AE, BE) using Claude on Amazon Bedrock. Auto-generates practice ladders, provides plain-language explanations and worked examples, and includes a Python + AWS Textract pipeline that ingests scanned exam PDFs into reviewable question drafts. Monetized via M-Pesa Daraja subscriptions.

Built an automated document formatting tool for Kenyan legal professionals, enabling court-compliant line numbering for pleadings, affidavits, and submissions. Integrated M-Pesa payments and supports High Court and Magistrate Court formatting standards per Kenya Civil Procedure Rules.

Built an LLM-powered legal research assistant (Claude + RAG over Kenyan case law) that helps litigators identify relevant precedents and draft arguments, reducing research time significantly. Currently serving 349 paying monthly subscribers.

Architected and built a system to digitize business operations, eliminating manual paperwork. Still in active use by the company.

Organized virtual and in-person technical events impacting students globally. Led the IoT team using Azure IoT, Arduino, and Raspberry Pi.

Led technical training sessions across multiple technology domains.