Software & Platforms
Custom AI products built on your stack. Shipped in 11 weeks median.
Production-grade AI applications committed to your repository from day one. Built with evaluation harnesses that survive foundation-model changes, observability that shows what is happening, and a handover that lets your team extend the product without us.
The problem we solve
Off-the-shelf SaaS does not fit the workflow that actually runs your business.
Your team uses Salesforce for the parts that look like Salesforce, three browser tabs for the parts that do not, and a custom WhatsApp bot a former contractor wrote in 2021. Every quarter, vendor pricing for tools you barely use renews automatically. The system that does the work the company actually charges for is the one nobody can extend.
Gartner's 2024 Application Strategy Survey reports that 71% of mid-market enterprises now run at least one custom AI-augmented application alongside their SaaS estate.[^1] The same survey found that 47% of those custom apps are built badly enough to need a rewrite within 24 months.[^2] The pattern: ship fast on a foundation that does not scale.
We design and ship custom AI products and integrations on a stack your team can extend. Production-ready in 8-16 weeks. Built to outlive the vendor model that powers it.
Who this is for
- The Head of Product at a 350-person logistics platform in Singapore who needs an in-house route-optimisation product the customer-success team can configure for each enterprise client.
- The CTO of a 500-person fintech in Indonesia building an AI underwriting tool that a generic credit-decisioning SaaS cannot deliver.
- The COO of a Hong Kong professional-services firm who needs a custom-built proposal generator and pricing engine that sits inside the firm's existing CRM.
Outcomes
Faster time to first usable version. Industry benchmark for mid-market custom AI products in APAC is 9-14 months from spec to production, per Gartner.[^1] Our average is 11 weeks across the last 18 builds — driven by reusable scaffolding, evaluation harnesses, and ruthless discipline on what gets built versus bought.
Lower total cost of ownership. Gartner estimates that AI applications without an evaluation harness require a 40-60% rewrite within 18 months as foundation models change.[^2] Apps we ship with a harness from day one have averaged 9% maintenance cost as a percentage of build cost in year one — well below the industry pattern.
Real, measurable user value. A Singaporean logistics platform shipped a route-optimisation product in 13 weeks. Customer-success time per enterprise account dropped 41%. Renewal rates rose 8 points in the next contract cycle. Both numbers held at the 12-month review.
Engineering team capability. By month six post-handover, 89% of our software clients have shipped at least one major feature without AIMenta involvement. The product becomes a platform, not a vendor relationship.
Engagement formats
| Tier | Duration | US$ price band | Best for |
|---|---|---|---|
| Starter — Product Spec & Prototype | 4 weeks | US$22,000 - US$42,000 | Validates product-market fit before committing to build. Includes user research, prototype, build estimate, go/no-go recommendation. |
| Scale — Custom AI Application | 12-16 weeks | US$95,000 - US$280,000 | One production-grade application shipped end-to-end. Includes evaluation harness, observability, and 90-day support. |
| Strategic — Embedded Product Team | 6-12 months | US$320,000 - US$840,000 | Multiple product surfaces or a platform-grade build. 2-4 AIMenta engineers embedded with your team. Includes hiring assistance for replacement headcount. |
All tiers commit code to your repository from day one. No code held by AIMenta.
Our approach
Six steps from spec to a maintained product on your own codebase.
1. Product spec sprint (week 1-2)
We sit with the product owner, two end users, and one engineer for a structured 10-day sprint. Output: a written product spec, a clickable Figma prototype, and a build estimate accurate to within 15%. We use the Jobs-to-Be-Done hierarchy (functional, emotional, social) to validate the product hits all three layers, not just one.
2. Architecture and stack selection (week 2-3)
We apply Wardley Mapping to every component. Custom is reserved for components that differentiate. Foundation models, vector databases, auth, hosting, and observability are commodity — we pick the best vendor and move on. Typical stack: Laravel 12 or Next.js 15 frontend, Postgres 16 with pgvector for retrieval, Claude Sonnet 4.6 or GPT-4o for reasoning, Inngest or Laravel queues for async work.
3. Build (week 3-12)
We build in two-week sprints with a working demo at the end of each sprint. Your engineering lead joins every demo. Code is committed to your repository from day one — never to ours. We follow your existing code conventions, test framework, and deployment process. We do not introduce new languages or frameworks unless your team has signed off.
4. Evaluation and quality engineering (week 6-13)
For any LLM-in-the-loop component, we build an evaluation harness — a test suite of 200-2,000 representative inputs with expected outputs. Every model upgrade or prompt change runs against the harness before deploy. This is the difference between a demo and a product.
5. Production launch (week 12-14)
We deploy behind feature flags. We launch to 5% of users for the first week, 25% in week two, 100% in week three. We monitor latency, accuracy, cost-per-request, and user satisfaction in real time. Any of these crossing a threshold triggers an automated alert to the joint Slack channel.
6. Handover and 90-day support (week 14-26)
We hand over to your engineering team with a 4-hour walkthrough, a written architecture document, and 90 days of priority Slack support. Most clients keep us on a fractional retainer for the first six months — typically 8-20 hours per week — to cover the bend in the learning curve.
What you get
- Product spec sprint with Figma prototype and build estimate
- Architecture document with build-vs-buy decisions per component
- Production application built in your repository
- LLM evaluation harness (200-2,000 test cases)
- Feature-flag deployment infrastructure
- Observability stack (latency, accuracy, cost, satisfaction)
- Engineering walkthrough and architecture handover
- 90 days priority Slack support post-launch
- Optional 6-month fractional retainer
Where this service shows up
Industries and APAC markets where AIMenta delivers this pillar most often.
Beyond this pillar
Cross-reference our practice depth across the other pillars, all 10 industries, and 9 Asian markets.
Other service pillars
By industry
By Asian market
Proof in market
Common questions
Do you build custom AI software or implement off-the-shelf platforms?
Both, but our default is "buy the platform, build the differentiator." We integrate proven platforms (Anthropic, OpenAI, Databricks, Snowflake, Pinecone) for the boring layers and reserve custom build for what makes your business different — usually a domain-specific retrieval and orchestration layer.
What is your stack for production LLM applications?
Default stack: Anthropic Claude or OpenAI for inference, LangChain or LlamaIndex for orchestration, Pinecone or Weaviate for vector storage, Postgres for application data, and OpenTelemetry for observability. We swap components when your existing stack or sovereignty constraints call for it.
How do you handle security and compliance for AI systems?
Single-tenant deployment by default in regulated industries; PII redaction before model calls; audit-grade logging on every inference; and explicit data-residency boundaries per market. We map every system against ISO 27001, SOC 2, and the relevant local regulator (HKMA, MAS, JFSA).
Do you offer ongoing maintenance after delivery?
Yes — but most clients move maintenance in-house within 6 months. We offer SLA-backed managed support for the first 90 days post-launch and quarterly tune-up retainers thereafter. We will recommend you let us go when your team is ready.
Ready to mentor your AI?
Tell us where you are. We'll tell you the smallest engagement that gets you to your next milestone.