When a redesigned workflow needs real software, Fetch builds it — from AI agents and data platforms to operational systems and custom products.
Engineering delivered for a national defence organisation, a national broadcaster, regulated financial issuers.
Headquartered in Singapore with engineering hubs in Hanoi and Ho Chi Minh City. Our engineering clients have included a national defence organisation, a national media broadcaster, regulated digital-asset issuers, an ed-tech operator and family-run businesses modernising their operations.
Agentic AI, real-time voice AI, institutional ledgers, wearable-data platforms, edge-to-cloud IoT — most of it in production today, not in a lab.
Every build starts with a clickable mockup of your product and a PRD, iterated with you until sign-off. The proposal is written against what you approved.
AI agents draft, test, review and document inside a process our senior engineers design and gate. It changes the speed you experience; it does not change who is responsible.
Documentation, runbooks and test cases are produced alongside the code. When a fixed-scope engagement ends you receive a repository your team can operate.
Open a practice to see what it has shipped to production — described by capability, not by client name.
AI that does a job, not a chatbot demo.
LLM-powered products end-to-end: model orchestration, retrieval, evaluation, guardrails and the plain product engineering around them. This is the engineering that turns a Sprint's redesigned workflow into software your team uses every day.
Enterprises have moved from “try an LLM” to “run a workflow on one”. The differentiator is no longer the model — it is orchestration, evaluation, cost control and provider independence.
| Multi-agent orchestration | LangGraph agent pipelines for content generation (brand DNA + campaign brief → a week of copy, image and video drafts) and creator tooling with a node-based canvas editor |
| Model-agnostic LLM gateway | A Go gateway abstracting OpenAI, Anthropic, Gemini and ~9 further providers behind one typed API — structured output, model listing, per-tenant keys — so a client is never locked to one model vendor |
| Real-time voice AI | Live, speaker-diarised transcription of video interviews from a Chrome extension, with the backend acting as a short-lived token broker so secrets never ship to the browser |
| Document AI | Scanned exam papers → OCR → LLM / vision extraction → marking schemes → human reviewer UI, versioned assessment pipeline |
| AI scoring & ranking | Candidate-scoring service with per-role rule configs, replacing a classic-ML stack |
| Conversational bots at scale | Telegram bot for a national broadcaster on Azure OpenAI with separate deployments for reply, moderation and engagement, queued on Redis, CMS-managed |
| Meeting intelligence | AI assistant that joins and records calls and produces briefs — the tool our own team runs on |
| On-device ML | In-browser face analysis with MediaPipe, paired with LLM interpretation server-side |
The systems a business runs on — built to be operated, not just launched.
When the workflow redesign needs more than a configured tool, we build the platform: catalogue to dispatch, back office to payments, CMS to storefront.
Most operational pain sits in the gaps between systems. We build the connective tissue — and the admin tools that let your team run it without us.
| Operations platform | End-to-end ops platform for a fresh-produce retailer: catalogue and inventory → orders → dispatch → rider delivery. Admin web, customer storefront, rider app, NestJS backend; payment-provider cutover absorbed mid-programme |
| Back-office & payments | Back-office with Stripe payments, S3 documents and JWE-secured APIs; admin SPA |
| CMS & marketing sites | Headless CMS admin for a chat-bot product; corporate websites with small auth APIs |
| Ed-tech SaaS | AI mathematics tutoring platform with subscriptions, question bank, exam papers, transactional email and analytics; Kubernetes with GitOps |
| Content platforms | AI campaign-content engine and a two-sided brand-contest platform for AI video creators |
Data platforms that ingest from the physical world and turn it into decisions.
Wearables, sensors and scales — with the identity, consent, ingestion and aggregation done properly.
Wearables have crossed from gadgets into institutional programmes — defence, insurers, employers — and edge-plus-cloud patterns let small operators get plant-floor data into the cloud without a SCADA budget.
| Wearable data platform | Fitness and health data platform for a national defence organisation: ingestion from consumer wearables → aggregation → API and two web applications (participant and admin), national digital-identity (SingPass OIDC) login |
| Corporate wellness | Wearable fitness-tracking platform with roles, leaderboards and device integrations, now under a security-and-scale upgrade |
| Industrial IoT / edge | Cloud weighbridge: industrial scale → Raspberry Pi edge service (serial capture, offline queue, retry on reconnect) → realtime cloud backbone → capture / review / admin web and an OpenAPI ERP interface |
Digital-asset infrastructure for regulated and institutional use.
For financial institutions and licensed issuers: privacy-preserving ledgers, stablecoin issuance and on-ledger settlement, delivered to production.
Stablecoin legislation in the US, MiCA in the EU and MAS's framework in Singapore have moved tokenised money from pilots into regulated production. We are one of the few Asia-Pacific teams with production DAML delivery.
| Canton Network / DAML | DAML smart contracts on the Splice / Amulet token standard; validator and participant operation; CIP-56 fungible-token implementation with delivery-versus-payment and a compliance blacklist extension |
| Institutional exchange | Partner / liquidity-provider swap exchange on Canton: on-ledger atomic-swap escrow (84 scenario tests), partner API, indexing and fee crons, wallet provider, TypeScript trading SDK — live on mainnet |
| Stablecoin issuance | Three generations for one licensed issuer, from an EVM pilot to a Canton-based issuance and tokenisation console driving the full mint / burn / transfer lifecycle |
| Custody & treasury | Hot-wallet token-manager and withdrawal-manager microservices, cross-chain settlement, multi-tenant deployments |
Client names are withheld under NDA; each engagement is described by sector and year. References can be arranged on request.
Chrome extension streams interview audio for diarised transcription; a scoring service applies per-role rules and LLM evaluation. Handed over to the client's own cloud in June 2026 and in production.
Spreadsheets and chat replaced by one system from inventory to the rider's phone, delivered fixed-scope in a 16-week programme. Handed over August 2026 with a full handover pack; phase 2 scoped.
Wearable ingestion → aggregation → participant and admin apps, SingPass OIDC login, dual CI with SAST and secret detection on every push. Fully containerised, no secrets in source.
Ten layers, primary choices and what else is live. Not a wish-list.
| Layer | Primary |
|---|---|
| Backend | TypeScript — NestJS, Prisma, Node 22 |
| Web | Next.js, React, Vite SPAs |
| Mobile | React Native |
| AI / ML | LangGraph, Anthropic, OpenAI, Azure OpenAI, Gemini; multi-provider gateway |
| Ledgers | DAML / Canton (Splice, CIP-56) |
| Data | PostgreSQL, Redis, BullMQ |
| Identity & payments | Auth0, SingPass OIDC, JWT / JWE |
| Cloud & delivery | AWS EKS + ArgoCD (GitOps), Terraform |
| CI & quality | GitHub Actions, Docker, Helm |
| Messaging | Telegram, WhatsApp, web push |
