About Me

I’m an Enterprise Architect at Akamai Technologies with over 15 years of experience building and scaling digital systems across mobile engineering, edge computing, application security and agentic AI. Lately most of my work goes into one question: how small can a capable language model get, and how close to the user can it run? I optimize purpose-built models and port them onto constrained serverless runtimes like Akamai Functions, where memory, cold starts and CPU-only execution decide what is possible.

Edge AI and model optimization for constrained runtimes

Most AI inference runs on GPU clusters. My focus is the opposite end: making models small and efficient enough to run inside WebAssembly-based edge functions, close to the user, with no GPU.

  • Model porting and optimization: Using distillation, quantization and architecture-level changes to fit transformer models within the memory, startup-time and compute limits of Akamai Functions.
  • Language at the edge: Indic transliteration and translation, plus language identification for Hinglish and romanized Indian languages.
  • AI guardrails at the edge: PII detection and redaction (including Indian identifiers such as Aadhaar and PAN), image moderation with OCR, audio transcription and document text extraction, all feeding into Akamai Firewall for AI.
  • Voice and vision workloads: Speaker verification to lock an AI assistant to one enrolled voice, and image conversion and resizing at the edge.
  • Research: Exploring transformer architecture changes designed for WASM execution, with the goal of publishing the technique.

Agentic AI, LLM Infrastructure & Real-Time Intelligence

  • Multi-Agent Systems & MCP: Architecting autonomous enterprise workflows using LangGraph, LangChain, and Model Context Protocol (MCP) architectures.
  • Edge AI & Guardrails: Deploying custom Small Language Models (SLMs) via WebAssembly (WASM) on Akamai Functions (Fermyon), and designing AI guardrail architectures inside Akamai EdgeWorkers using Firewall for AI.
  • Inference, RAG & Voice Pipelines: Fine-tuning LLMs, building domain-specific RAG pipelines, self-hosting inference stacks (vLLM, Ollama) on Linode, and engineering localized, open-source speech-to-speech (S2S) audio pipelines.
  • Agent Infrastructure & Governance: Building Telegram-controlled AI agents, local code-execution sandboxes, and authoring proposals for east-west agent control planes.

Mobile Engineering, SDK Scale & Deep Security

  • Production Scale & SDK Engineering: Launched 45+ applications on the App Store and Google Play (iOS, Android, Flutter, React Native). Resolving production root causes at scale in applications, including memory bugs, thread-exhaustion OOMs, and native library load failures.
  • Offensive & Defensive Security: Performing Frida-based dynamic instrumentation, reverse engineering, smali bytecode analysis, and pentesting to harden enterprise mobile applications.

Edge Architecture & High-Concurrency Systems

  • Edge-Native Runtimes: Integrating database-fused edge runtimes (Akamai EdgeWorkers, HarperDB, Fermyon Spin) with global CDN infrastructure for high-volume WebSockets and real-time traffic.
  • AI Security Assessments: Auditing production delivery configurations and evaluating trust assumptions in AI-integrated customer architectures.

Certifications

  • AWS Certified Solutions Architect – Associate
  • CIW JavaScript Specialist

Recognition

  • Akamai Global Services “Most Impactful Person” award, 2026

I’m focused on the responsible adoption of AI-driven architectures, and especially on showing that useful AI doesn’t always need a data center behind it. I share what I learn on this blog across mobile security, edge engineering and edge AI, and I enjoy mentoring engineers who are working on the same problems.

by
Fareeth John