// Singapore · Agentic Engineering House

Drop-in agentic workflows for the stack you already run.

We build, deploy, and operate AI agents inside your existing tools — email, CRM, drive, ticketing. No platform migration. No workflow rebuild. Your team keeps working the way they already work; the agents do the rest.

Start a conversation → Watch how it works ↓

Inside the Workflow — 45s look at how a Qlabs agent slots into an existing team's tooling.

// What we build

Agent-native engineering, end to end.

From the first agent that reads your inbox to the platform that operates them at scale — without your team changing how they already work.

01

Agentic Workflows

Multi-step AI agents that read your existing tools — inbox, CRM, drive, ticketing, Slack — and act inside them. We design the agent around your process, not the other way around. No platform migration, no rip-and-replace, no behaviour change for the team.

02

AI-Integrated Systems

LLM features wired directly into the apps your team already uses, via Anthropic Claude and Qwen. Surface intelligence where work already happens — no new dashboard to log into.

03

Workforce Augmentation

Staff-facing AI surfaces that take rote work off the team's plate and free human time for judgment-heavy work. Built around the Automate / Augment / Amplify / Anchor framework and aligned to Singapore grant evidence requirements.

04

Full-Stack Platforms

When a custom system is the right answer: React frontends, FastAPI and Django backends, PostgreSQL, Redis, Docker, deployed to AWS.

05

Cloud Infrastructure

AWS architecture across EC2, S3, RDS, SES, and Route 53 — security-hardened, cost-optimised, and production-ready.

06

Ops Automation

Internal platforms that replace spreadsheets and off-the-shelf tools with purpose-built systems for scheduling, billing, and reporting.

// Selected case studies

Agentic AI, shipped.

Three production systems where the agent isn't a chat window bolted on — it runs the workflow. Tool-calling, RAG, output validation, persistent memory, and multi-provider LLM abstraction, all behind real surfaces in real businesses.

Guru Confidential — under NDA

AI coaching co-pilot for financial planners.

Closes the loop between client meetings and coached follow-ups. Planners record a debrief; Guru runs a structured agent pipeline that diagnoses the conversation, retrieves the right framework for the situation, and ships a validated coached report — Telegram chat or PDF — in minutes.

Telegram bot · FastAPI web dashboard · PDF reports

// Agentic capabilities

  • Multi-step agent pipeline (parse → diagnose → retrieve → generate → validate → regenerate) with enforced section completion
  • Weighted RAG over expert frameworks (50% challenge / 30% archetype / 20% guru), tuned by user-confidence signal
  • Multi-provider LLM abstraction — Anthropic, Gemini, Qwen, Ollama swappable via config
  • Visual diagnostic scorecards (six 1–10 metrics + closing-probability prediction) rendered to PDF via headless Chromium
  • Persistent per-user memory: navigation, chat history, meeting materials within a 50k-char context budget

// Stack

pythonfastapianthropicsqlalchemyplaywrightdocker
Insurance CRM Confidential — under NDA

Self-hosted CRM for independent insurance agents.

Replaces spreadsheets and manual policy entry. Drop in a policy PDF or paste raw insurer-portal text; the system extracts 40+ structured fields, classifies status, schedules reviews, and tracks ILP performance — multi-tenant, local-first, no cloud lock-in.

Self-hosted web app · multi-user local-first · WhatsApp sync

// Agentic capabilities

  • LLM document extraction (Qwen text + vision fallback) mapping PDFs to 40+ policy fields including riders, exclusions, and coverage
  • Pre-LLM regex + fuzzy-alias parser ("Smart Paste") reduces hallucination on structured insurer text
  • Settings-driven LLM config — swap providers and base URLs per tenant without code changes
  • Graceful degradation: semantic / keyword extraction fallback when LLM is unavailable
  • Automated ILP TVM checks, policy review scheduling, and client appreciation calendar

// Stack

pythonfastapihtmxalpine.jssqlalchemyapschedulerqwen
Lion City Exams Live · lioncityexams.com ↗

AI-augmented tutor marketplace for Singapore.

Two-sided marketplace matching parents with verified tutors. Claude ranks candidates in real time, an autonomous Telegram agent fields subject questions inside a 400+ member community, and an orchestrated email-relay workflow takes a shortlist all the way to paid lesson — without a human coordinator in the loop.

Web SPA · Telegram bot · email-relay · Stripe payments

// Agentic capabilities

  • Claude-driven matching: ranks 20+ candidate tutors against parent requirements in under a second, returning top 3 with personalised explanations; rule-based fallback when the LLM is unavailable
  • Autonomous Telegram agent monitors a 400+ member group for subject/level keywords, waits 15 minutes for human responses, then auto-posts tutor cards
  • Email-relay agent (Claude + Gmail API) drafts and mediates tutor ↔ parent conversations — replies via signed links, no auth, no direct contact until payment clears
  • Time-triggered nurture orchestration via APScheduler: shortlist nudges (+24h / +72h / +7d), commission retries, lesson-confirmation prompts, review-unlock check-ins
  • Review-gated quality system: lessons must be confirmed as occurred before reviews unlock; tutor quality score auto-recomputes from base + review bonus

// Stack

pythonfastapianthropicreactstripegmail-apiapschedulerdocker
MindlabOps Confidential — under NDA

Agentic operations platform for an education business.

One platform replacing Mindbody and 15 spreadsheets across enrolments, scheduling, finance, and communications for a Singapore tutoring centre. Non-technical staff drive operations through a permissioned chat agent that books trials, records attendance, and processes invoices under explicit human confirmation.

Internal web app · scheduled Celery workflows · staff chat agent

// Agentic capabilities

  • Tool-calling agent with 40+ tools across 10 domains (students, scheduling, finance, attendance, tutors, marketing)
  • Role-based tool access — admin, finance, marketing, tutor, readonly — enforced per call
  • Confirmation-first design: agent clarifies ambiguity and asks before any write operation
  • Persistent agent memory — staff teach the system standing rules, stored in Postgres and injected per session
  • Multi-backend LLM (Ollama / Anthropic / DashScope) behind a single interface; conversation history persisted for continuity

// Stack

djangopostgrespgvectorredisceleryhtmxanthropicdocker

// Selected work

Products shipped to production

Quivaro quivaro.com ↗
FastAPIReactPostgreSQLS3

Internal platform for international curriculum specialists. AI-powered worksheet generation, 18,721 IGCSE and IB study resources, six-hourly S3 backups.

OpenClaw Internal
PythonAWSClaudeQwen

Autonomous agent platform on AWS Lightsail — email watch, enquiry routing, Google Drive integration, dual-LLM architecture with Anthropic Claude and Qwen.

Workforce Intelligence Platform Confidential
DjangopgvectorClaudeOpenClaw

Job redesign programme and five staff-facing AI surfaces for a 170K-member Singapore digital media platform — AEO intelligence, prospecting assistant, account health monitor, and moderation copilot, with a workforce-analytics dashboard tracking hours reallocated, skills-assessment closure, and revenue-per-FTE delta.

// Insights

Thinking on AI and workforce

All articles →

// Get in touch

Let's talk about your project.

Based in Singapore. Working with teams globally.

hello@qlabshq.com →