Southampton · --:--

Kyle Bown.

Full-Stack Developer

I design and build end-to-end web platforms — from the interface people touch down to the distributed infrastructure underneath it. Currently Software Manager at GoBig Software Group, leading development across Gotalk.ai and GoSafe.ai.

About

Experienced full-stack developer with a strong background in both front-end and back-end engineering — and a habit of owning the whole lifecycle, from architecture through to deployment and tuning.

Based in Southampton, UK, I work on designing and developing responsive, scalable web applications while contributing to successful projects at GoBig Software Group, where the products are white-labelled by resellers, telecom providers and MSPs. My day-to-day spans architecture, leading the development team, and keeping delivery on schedule.

The work runs the full width of the stack: interface design and front-end build, API and database architecture, AI and voice integration, and the cloud clusters and virtual data centres that keep it all running.

3

Platforms led end-to-end

<100ms

Breach lookup latency

Bn+

Records indexed & searchable

Selected work

Platforms I've architected, built and shipped.

Three products, one common thread — own the architecture, then own the delivery.

01 — Security & data infrastructure

GoSafe.ai

Software Manager2025 — Present

An enterprise breach-intelligence and dark-web monitoring platform: a multi-tenant PHP application for customers and white-label distributors, sitting on a purpose-built distributed data platform that ingests and searches breach corpora at billions-of-rows scale.

  • Built the ingestion platform end-to-end — FastAPI and a Python worker fleet with Rust extensions (PyO3) that schema-detect an unknown dump at 50–100 MB/s.
  • Designed the store: a replicated columnar cluster — one shard across three replicas under Raft consensus, partitioned 64 ways, tiering from hot disk out to cold object storage at 90% capacity.
  • Associations are AES-256 encrypted at rest and keyed by SHA-256 hash, so lookups resolve without ever decrypting the corpus.
  • Shipped the customer platform: distributor / account / user multi-tenancy with white-label branding, domain and IP monitoring, lookalike-domain detection, phishing simulation, incident response and a key-authenticated public API.
Rust Python FastAPI PHP Redis Docker Swarm
Visit go-safe.ai
GoSafe dashboard showing dark-web monitoring, breach breakdown and risk scoring

02 — AI & voice technology

Gotalk.ai

Software Manager2023 — Present

An AI text-to-speech platform for on-hold marketing and IVR. A multi-tenant PHP application with domain-routed white-labelling, sitting on a containerised Python voice service that handles synthesis, translation, phonemes and audio mixing.

  • Built the voice service — a containerised Python API behind a multi-worker pool, with an offline translation sidecar, phoneme generation and FFmpeg audio processing.
  • Designed segment-level caching keyed by content hash, so an edited script only re-synthesises the segments that actually changed.
  • Multi-voice SSML with automatic voice switching, plus auto-ducking that mixes speech over music through a fallback chain rather than failing the render.
  • Added an operations console with cluster-wide cache kill switches that apply without a redeploy, a circuit breaker, per-IP rate limiting and a contract test suite that pins the public API against a live container.
Python Flask PHP FFmpeg SSML Docker Swarm
Visit gotalk.ai
Gotalk.ai project dashboard and voiceover timeline editor

03 — Platform group & team leadership

GoBig Software Group

Software Manager2023 — Present

GoBig builds its own SaaS platforms and takes them to market through a partner channel — resellers, telecom providers and MSPs who white-label the products as their own. I lead development across the group: the platforms themselves, and the shared foundation every one of them runs on.

  • Lead development across the product portfolio, including GoSafe and Gotalk.ai — setting the architecture, steering the roadmap and owning delivery.
  • Built the shared foundation the products sit on: centralised identity, a domain-routed multi-tenant core that serves every brand from one codebase, and the billing and agreement flows behind partner accounts.
  • Made white-labelling a first-class concern — each partner gets their own domain, branding and isolated customer base, so a reseller's customers only ever see the reseller.
  • Work directly with partners to turn channel requirements into product, rather than a trail of one-off customisations.
Platform architecture Multi-tenancy White-label PHP MySQL Team leadership
Visit gobigsoftware.co.uk
GoBig Software Group logo

Under the hood — GoSafe

How GoSafe holds billions of records.

Two halves of the same system — how a raw breach dump gets in, and how a lookup gets back out fast.

Write path — ingestion

Breach dumps arrive as whatever the leaker happened to export. The pipeline's job is to turn that into something a query can trust — without ever leaving the data readable.

  1. Detect the schema

    A Rust extension bound into Python through PyO3 works out the shape of an unknown dump at 50–100 MB/s. SQL, JSON/NDJSON, CSV/TSV, XML, combo lists and raw text all enter the same funnel.

  2. Map and extract

    An operator confirms the field mapping in the React console. SQL is split per table by the Rust parser and extracted in memory, so no staging file ever hits disk.

  3. Encrypt before it lands

    Associations are AES-256 encrypted on the way in and emails are keyed by SHA-256 hash. The corpus stays searchable without being decrypted, so a lookup never handles plaintext.

  4. Write in batches

    Records land in the store as 500,000-row columnar batches. Work is dispatched to a worker fleet through a Redis queue, with live progress streaming back over WebSockets.

  5. Replicate and tier

    One shard across three replicas, with Raft handling consensus. Emails and associations are aggregated and replicated across 64 partitions, tiering out to cold object storage once a node passes 90% capacity.

GoSafe breach ingestion architecture Breach dumps in SQL, JSON, CSV, XML and combo-list formats are schema-detected by a Rust parser, extracted in memory, AES-256 encrypted and written in columnar batches by a Redis-dispatched worker fleet into a three-replica columnar cluster partitioned 64 ways, which tiers from hot local disk out to cold object storage at 90 percent capacity. BREACH SOURCES sql · json · csv · xml · combo lists RUST PARSER · PyO3 schema detection 50–100 MB/s DISTRIBUTED WORKERS redis queue · cpu_count − 1 per node extract aes-256 columnar batch REPLICATED COLUMNAR STORE 1 shard × 3 replicas · raft consensus replica 01 replica 02 replica 03 64 partitions · replicated + aggregated TIERED STORAGE hot disk → cold object storage at 90% searchable, never decrypted

Read path — try it

Watch a lookup prune 64 partitions.

An illustration of the query plan — not a live search. It runs entirely in your browser, queries nothing, and no data is sent anywhere.

    Under the hood — Gotalk

    Why an edited script costs almost nothing.

    Synthesis is the expensive part of a voice platform. The trick is making sure you only ever pay for it once.

    Request path — synthesis

    A script is rarely written once. It gets tweaked, re-recorded, a name changed, a line dropped — and a naive pipeline re-synthesises the whole thing every time.

    1. Split the script

      An SSML script can switch voice mid-sentence and carry breaks, prosody and phoneme overrides. The segmenter breaks it into per-voice segments and fingerprints each one, so a segment is the unit of work — not the request.

    2. Look up each segment

      Every segment is checked against object storage by content hash. Because the cache is keyed per segment rather than per response, an untouched sentence in an edited script is never synthesised twice.

    3. Synthesise only the misses

      Only segments with no cache entry reach the speech provider; the rest are served straight from storage. That is what keeps synthesis spend proportional to what actually changed, rather than to how often someone hits render.

    4. Mix and render

      FFmpeg handles format conversion, and auto-ducking mixes speech over background music through a fallback chain — a failed analysis degrades to a simpler mix instead of failing the render outright.

    5. Keep it standing up

      A circuit breaker, per-IP rate limiting and host allow-listing sit in front of the service. Cache reads and writes can be killed cluster-wide from an operations console without a redeploy, and a contract suite pins the public API's observable behaviour so it can be refactored without breaking consumers.

    Gotalk speech synthesis request path An SSML script is split into per-voice segments, each fingerprinted by content hash. Every segment is looked up in object storage: cache hits are served directly, while misses go to the speech provider and are stored on the way back. Both paths merge into an FFmpeg mixing stage that handles ducking, conversion and normalisation, so only changed segments ever cost a synthesis call. SCRIPT + SSML multi-voice · breaks · prosody · phonemes SEGMENTER per-voice split sha-256 per segment CACHE LOOKUP per segment, not per request hit miss OBJECT STORE cached segment TTS PROVIDER synthesise + store MIX & RENDER ffmpeg · auto-duck fallback chain duck convert normalise changed segments only

    Segment cache — try it

    Render it twice. Then change one line.

    An illustration of the caching model — no speech is synthesised. It runs entirely in your browser, calls nothing, and no data is sent anywhere.

    cache empty

      Capabilities

      What I bring to a project.

      Front-end craft, back-end depth, and the infrastructure knowledge to keep both fast under load.

      Full-Stack Development

      • End-to-end web solutions across front-end (React, HTML, CSS, JavaScript) and back-end (Node.js, Express).
      • Database management with MongoDB, MySQL and PostgreSQL.
      • RESTful APIs and GraphQL query systems for scalable, efficient data handling.
      • Cloud deployment on AWS and Azure, with GitHub versioning into production.

      UI/UX Design & Creative Direction

      • Seamless, responsive and accessible interfaces for platforms like Gotalk.ai.
      • Brand strategies, design systems and high-fidelity prototypes built for developer hand-off.
      • User-flow wireframing and customer-journey optimisation for real usability gains.

      AI Integration & Voice Technology

      • Streamlined text-to-speech workflows at Gotalk.ai, improving voice generation and IVR systems.
      • Machine-learning API integration for speech processing and dynamic content.
      • Automating and optimising creative processes with AI-driven tooling.

      Cloud Clusters & Virtual Data Centres

      • Designing, deploying and managing clusters with Kubernetes and Docker Swarm.
      • Virtual data centre management — provisioning, workload optimisation, high availability.
      • Hybrid environments integrating on-premises infrastructure with cloud.
      • Monitoring with AWS CloudWatch, Azure Monitor and Prometheus.

      Tools & Platforms

      • Figma, Adobe Creative Suite, and user-testing tools including Google Analytics.
      • CI/CD pipelines using Docker and Git workflows.
      • CRM integration with HubSpot and Mailchimp for customer engagement.
      • Local server setups with EasyPHP, versioned through to production.

      Technical Leadership

      • Leading development teams and managing project timelines end-to-end.
      • Translating partner and channel requirements into deliverable technical scope.
      • Owning the full lifecycle — architecture, deployment, scaling and continuous tuning.

      Toolkit

      The stack I reach for.

      Front-end

      ReactTypeScriptJavaScriptViteTailwind CSSHTMLCSS

      Back-end

      Node.jsExpressPHPPythonFastAPIFlaskRustFFmpegRESTGraphQL

      Data

      MySQLPostgreSQLMongoDBRedisSQLiteColumnar analyticsDistributed query

      Cloud & infra

      AWSAzureDockerDocker SwarmKubernetesGlusterFSMinIO / S3CI/CD

      Observability

      GrafanaPrometheusCloudWatchAzure Monitor

      Design & growth

      FigmaAdobe Creative SuiteGoogle AnalyticsHubSpotMailchimp

      Contact

      Talk to me about
      the hard parts.

      Architecture decisions, systems at awkward scale, or something under the hood that isn't behaving. Those are the conversations worth having.