The Stixis AI Development Life Cycle

Six phases. Six gates.

Every use case, from a first MVP to agents in production, runs on one life cycle. Nothing moves to the next phase until its gate is signed.

Why one life cycle

Speed you can stand behind.

AI projects stall in two places: before anything is built, when nobody agrees on value and risk, and after the demo, when nobody has proved it is safe to depend on. The AIDLC fixes both with a gate at each step.

The same phases, outputs and gates apply whether you buy a fixed-bid MVP, a retained pod or a full capability centre, so work moves between teams without starting over.

The six phases

What happens, what you get, what must be signed.

01

Discovery

Frame the problem, the value and the risk before anything is built.

  • Use case brief with baseline and target
  • Risk tier (1 to 4) and autonomy level (A0 to A4)
  • Value hypothesis and success measures
Gate 1 · Use case approvalValue, feasibility and risk tier agreed by the business owner before data work starts.
02

Data Foundation

Make the data the use case needs reachable, clean and permitted.

  • Source inventory and access plan
  • Quality profile and fixes
  • Privacy and residency review
Gate 2 · Data readinessSources, quality and permissions signed off for the risk tier.
03

Model Selection & Tuning

Choose and tune the model against your own golden set, not a benchmark.

  • Golden evaluation set
  • Model comparison on accuracy and cost per interaction
  • Tuning or retrieval design
Gate 3 · Model choiceThe chosen model meets the agreed accuracy target at an acceptable cost.
04

Agent Orchestration

Wire models into workflows, tools and people.

  • Agent design with tool allowlist
  • Human checkpoints set by autonomy level
  • Integration with systems of record
Gate 4 · Design reviewEvery tool call, approval and fallback is designed and reviewed.
05

Guardrails & Evaluation

Prove it is safe and accurate before people depend on it.

  • Guardrails for privacy, safety and policy
  • Red-team test against the OWASP LLM Top 10
  • Evaluation report against the golden set
Gate 5 · Release readinessAccuracy target met, red-team findings closed, audit trail switched on.
06

Scaled Deployment

Run it in production and keep it working.

  • Monitoring for drift, cost and incidents
  • Runbooks and kill switch
  • Handover or managed operation
Gate 6 · OperateOwnership, monitoring and incident response confirmed for live use.
Guardrails by design

Every use case gets a risk tier and an autonomy level.

Set in Discovery and checked at every gate. The higher the tier, the stronger the evidence needed before release; the higher the autonomy, the more checkpoints a person keeps.

Risk tiers
T1Critical
T2Elevated
T3Standard
T4Minimal
Autonomy levels
A0Assist: suggests, a person acts
A1Draft: prepares, a person approves each output
A2Act on low-risk steps, people review
A3Act with approvals at set checkpoints
A4Autonomous within guardrails, with a kill switch

Controls map to NIST AI RMF, ISO/IEC 42001 and the EU AI Act. Agents are tested against the OWASP LLM Top 10 before release.

Commercial models

Buy it the way you work.

Start small with a fixed-price proof, scale with pods or managed operations, or build a capability centre you own. Rates come from one Stixis rate card, shared with each proposal; nothing on this page is a price.

ModelHow it works
Fixed BidDefined scope and outcome at a fixed fee; Stixis carries delivery risk.
Managed Capacity (T&M)Named specialists on time and materials, inside your team.
Retained AI PodA standing cross-functional pod on a monthly run rate, with fixed-bid outcomes.
Managed ServicesStixis runs it in production against agreed service levels.
India Technology Centre (BOT)A dedicated centre in Bangalore, built and run by Stixis, then transferred to you.
Managed Advisory RetainerA fixed monthly retainer with a named lead, after a three-month minimum.
Contract protections

The same terms, every time.

These sit in every proposal and statement of work, so you know where you stand before work starts.

IP ownership

Work product built for you is yours. Pre-existing Stixis tools are licensed to you for the engagement and at handover.

Data privacy

A data processing agreement, data kept where you need it, and no client data used to train shared models.

Hallucination SLA

For generative AI, accuracy is measured against your own golden set, with an agreed target and remedy.

Responsible AI

Every use case gets a risk tier and autonomy level, with human checkpoints and an audit trail.

Change control

Scope changes are written, costed and approved before work moves.

Where we deliver from

Three centres, one way of working.

Leadership and client teams in Dallas–Fort Worth, engineering depth in Bangalore, and delivery in Johannesburg, working to the same life cycle and standards.

Dallas–Fort Worth, United States
Stixis Technologies, Inc.
Crestwood Center, 1200 W Walnut Hill Ln, Suite 2050
Irving, TX 75038
Bangalore, India
Stixis Technologies Private Limited
#477, Urban Vault, 18th Cross Road, Parangipalya, Sector-2
HSR Layout, Bangalore 560102
Johannesburg, South Africa
Stixis Technologies, SA
Building No 8, Pinewood Square, Pinewood Office Park
33 Riley Road, Woodmead, Johannesburg 2191
Standards we work to
ISO 27001aligned practicesSOC 2aligned practicesCMMIaligned practicesISO 9001aligned practices

Your own centre in India: GCC build-operate-transfer

Pick one use case. We'll take it through gate one.

Tell us the process you want to change and we'll propose a Discovery plan.

Talk to an expert →