Discovery through deployment
7 stages. We own all of them.
Most AI engagements are scoped to the middle 3. The client discovers at handover that evaluation, deployment and operation were somebody else's job, and that somebody was never hired.
- 1
Scope
KPIs, baseline and test set agreed
Week 0
- 2
Architecture
Design, cost model and plan signed off
Week 1
- 3
Build
Working demo from sprint 2
Weeks 2 to 6
- 4
Evaluate
Scored against the benchmark
Week 7
- 5
Deploy
Into your cloud account
Weeks 8 to 10
- 6
Operate
Monitoring, SLAs, on-call
Week 11
- 7
Handover
Docs, runbook, code walkthrough
Week 12
1 accountable engineer
No account manager relaying messages between you and a team you never meet.
No subcontracting
The person who designs the architecture writes the code and runs the deployment.
No platform lock-in
There is no licensed product underneath. Nothing stops working if you stop paying us.
Handover is a deliverable
Documentation, runbook and a walkthrough with your engineers, beyond a repository link and goodbye.
How we start
From first call to a measured result in 14 days
The first 2 weeks are designed so you can stop at a pilot fee if the number is not there. Most clients do not stop.
Day 0
Scoping call
30 minutes with the engineer who will build it. If AI is the wrong tool for the problem, we say so here.
Days 1 to 2
Written scope
Success metric, architecture, what is out of scope, fixed price and dates, in writing.
Days 3 to 10
Baseline and first slice
Your evaluation set comes first, then a working slice on your real data that you can query.
Days 11 to 14
Scored result
The slice is scored against the agreed set. If the number is there, we scale it. If not, you stop at a pilot fee.
After launch
Run and improve
Monitoring and improvement cycles on retainer, or a clean handover to your team.
You commit budget 1 stage at a time
Stage 1 · 2 weeks
Pilot
1 workflow, 1 success metric, fixed price. You see a scored result on your own data before committing a larger budget.
Stage 2 · 2 to 16 weeks
Production build
The same architecture scaled out with integrations, guardrails, monitoring and security review, delivered in milestones you approve.
Stage 3 · 6 months and longer
Retainer or team
Monthly improvement cycles or model upgrades, or AI engineers placed inside your team, working in your time zone.
What you own at handover
- Full source with commit history
- Fine-tuned weights where applicable
- The evaluation set your team can re-run
- Architecture diagram and decision notes
- Runbook for the failure modes we expect
- Dashboards and alerts already wired
- Recorded handover plus 30 days of questions at no charge
What ships with every system
- An evaluation set with a numberBuilt from your data before the system, re-run on every change.
- GuardrailsPII redaction, policy checks and confidence gates on every output.
- Tracing and dashboardsLangfuse or LangSmith on every call, with quality, latency and cost in 1 view.
- Alerts that fire firstThresholds set from real traffic, so you hear about drift before a customer does.
- A runbookWhat to do when retrieval quality drops, a provider throttles you or a check fails.
- A rollback pathVersioned prompts, models and indexes, so any release can be reversed in minutes.
What end to end means in practice
The list procurement asks about, and where the industry default usually sits
Shorter builds compress the same shape into 2 to 4 weeks. What does not change is the order: KPIs before architecture, architecture before code, evaluation running alongside the build rather than at the end.
| Deliverable | Typical AI vendor | Jillani SofTech |
|---|---|---|
| Written architecture and trade-offs | Slide deck | Documented, in your repo |
| Success metrics agreed before code | Rarely | Week 0, reported weekly |
| Evaluation harness you can rerun | Not included | Included every time |
| Deployment into your cloud account | Often extra | Included |
| Monitoring, alerting and drift detection | Phase 2 | Live at launch |
| Source code and IP ownership | Licensed platform | Yours from commit 1 |
| Runbook and engineer walkthrough | Repo link | Documented handover |
| Post-launch support | New contract | 30 days minimum, retainer after |
How we de-risk a 6-figure engagement
Anyone can put percentages on a website. This is the machinery that makes ours checkable.
The metric comes first
The success metric, the baseline it has to beat and the test set it is scored on are agreed in scoping, before serious money moves.
You approve before we build
The documented architecture, the cost model and the phased budget are signed off before the first build sprint starts.
Working software from sprint 2
You judge a demo you can click rather than a status report, every single week.
An eval gate guards go-live
Nothing is promoted to production until it beats the scored test set agreed in scoping.
Your accounts from the first commit
Code, pipelines and infrastructure live in your repositories and your cloud, so nothing is ever held hostage.
Accountable after launch
Monitoring, SLAs and quarterly improvement cycles sit inside the engagement rather than behind an upsell.
Asynchronous by default
Written updates, recorded demos and a shared decision log. You can read where the project stands without waiting for a call, and so can anyone you forward it to.
1 fixed checkpoint a week
30 minutes, same slot, set to your calendar. Progress against the agreed metrics, blockers named plainly, and the decisions we need from you. If there is bad news it arrives here, well before handover.
A decision log that outlives the project
Every architectural choice is recorded with the alternatives considered and why they lost. 6 months later, when someone asks why the retriever works this way, the answer is written down.
What we ask from you
1 named decision-maker who can approve architecture and unblock access, and a subject expert available for a few hours in the first 2 weeks to build the evaluation benchmark.
Pricing and engagement models
Clear scope, clear pricing, no surprises at invoice time
4 ways to work together, 1 standard of delivery. Every engagement ends with a working system in your cloud, documentation your team can use and code you own outright. Final pricing is confirmed in writing after a free scoping call.
Architecture Review
For teams that want a written plan in hand before committing to a build.
from $2,500
2 weeks, fixed price
- Review of your data, current system and goals
- Written architecture with trade-offs
- Evaluation plan and success metrics
- Build cost and timeline estimate
- Yours to take to any vendor, including us
RAG or Automation Build
A focused system that solves 1 clear problem, shipped fast.
from $3,500
2 to 4 weeks
- RAG system or n8n workflow on your data
- Hybrid retrieval, reranking and citations
- Cloud deployment and documentation
- Evaluation against ground-truth benchmarks
- 30-day post-launch support
Enterprise AI Platform
A full product with agents, APIs, dashboards and governance.
$15K to $60K+
8 to 16 weeks
- Multi-agent systems and LLM SaaS
- Backend, auth, multi-tenancy and analytics
- Full MLOps and LLMOps pipeline
- HIPAA, GDPR or SOC 2 architecture
- Monitoring, SLAs and quarterly reviews
Retainer and Team Augmentation
Ongoing engineering capacity embedded in your team.
$4.8K to $6K
per month, 6 months or longer
- Dedicated AI engineering hours each month
- Model tuning and retrieval improvements
- Incident response and managed monitoring
- Architecture reviews and roadmap input
- Priority access to the engineer directly
What is in the price regardless of engagement size
These are the reason a system still works 6 months after handover, and none of them is an add-on.
What moves the number
2 projects described in the same sentence can differ by a factor of 5 in effort. These 5 variables usually explain the gap, and they are the ones we probe hardest on a scoping call.
- 1Data readinessClean, accessible, well-structured sources are the single biggest cost saver. Scanned PDFs, inconsistent schemas and undocumented legacy exports add weeks.
- 2Number of integrationsEach system the build has to read from or write to carries its own authentication, rate limits, error handling and testing.
- 3Compliance requirementsHIPAA, GDPR, SOC 2 and CSRD change the architecture as well as the paperwork, and that work happens up front.
- 4Accuracy targetMoving from a useful system to a system that can be relied on in client-facing work is usually where the last third of the effort goes.
- 5Operating modelWhether your team runs the system afterwards or we do changes the shape of the engagement and the ongoing cost.
The figures on this page are starting points. A real number comes after a scoping call where we have seen the data and the constraints, and it arrives in writing with the assumptions listed. You contract with Jillani SofTech, a registered IT services firm: MSA and NDA on request, USD invoicing, international bank transfer, and milestone payments agreed in writing before work starts.
Security and data handling
Written for the person reviewing you before contract
Deployment models, data residency, access control, retention, sub-processors and incident response. Full detail is available under NDA for procurement and security reviews.
Your cloud account
Every component is deployed into infrastructure you own and control, with billing you can see and region pinning where a regulator names a jurisdiction.
Data handling
Production data never enters development environments. Model providers are used under zero-retention terms, or replaced by open-weight models inside your VPC when nothing may leave.
Access control
Least-privilege roles, SSO where you have it, secrets in a managed vault, and access revoked at handover with a written confirmation.
Contracts
Master services agreement, NDA and data processing terms from the company on request. Sub-processors listed in writing before work starts.
Audit trail
Every model output traceable to the prompt version, model version and retrieved context that produced it. SOC 2 and HIPAA architectures delivered.
Incident response
Named contact, response targets agreed in the engagement, and a written post-incident summary for anything that touches production.
Questions procurement teams ask
What clients want settled before a contract is signed
The founder leads architecture and delivery on every engagement and writes a large share of the code. Where a project needs additional capacity we bring in engineers from the team under the same accountability. Work is never passed to an agency or subcontracted.