Applied AI Studio · Dhaka, Bangladesh
WedeployAIsystemsinsiderealoperations.
Kitalon Labs maps the workflow as it actually runs, proves the system with evals, and ships it on top of the software you already use. Sandboxed first. Logged always. Approved by humans where it matters.
What a run looks like: every action recorded, every step reviewable.
Position
Everyone can buy the same models. The advantage moved to deployment.
MIT reports that 95 percent of generative AI pilots fail. They rarely fail at the model. They fail at the workflow: the undocumented exceptions, the systems that never talk, the approvals nobody wrote down. We build for that layer.
That is forward deployed work. Sit inside the operation, find the steps where judgment is actually needed, and wire intelligence into those. Everything else stays deterministic software.
Method
Diagnose
Fixed-fee sprintWe sit with the people doing the work and map every step, exception, and handoff. You get an operating map and a ranked automation matrix: what to build, what to keep deterministic, what to leave alone. Useful even if you never hire us again.
Prove
Evals before productionYour real cases become a golden dataset. We test the system against it and publish the results: pass rates, failure categories, escalation behavior. Non-determinism becomes evidence you can approve.
Deploy & operate
Your stack, not a migrationThe system ships on top of the software you already run, from ERPs to spreadsheets. Sandbox, then shadow mode, then autonomy where it earns it. Every action logged. Humans approve what matters.
Then the loop runs again. Fixing one workflow exposes the next bottleneck upstream or downstream. That is why engagements compound instead of ending.
Engagements
The entry point. One workflow, mapped as it actually runs.
- Operating map of one workflow
- Ranked automation matrix
- Working prototype on your data
- Fixed fee, fixed scope
The full system, hardened for the thousand ways things go wrong.
- Structured outputs and failure handling
- Eval suite with golden dataset
- Complete audit trail
- Deployed into your existing stack
We run what we ship. Systems drift; someone has to watch.
- Monitoring and eval refresh
- Model cost tuning
- Exception review cadence
- Expansion when evidence supports it
Prices are anchors, not quotes. Final numbers follow the sprint.
Work
No logo wall. Each engagement below is checkable.
LEAD Academy
Career guidance system in production: assessments, skill-gap mapping, and course recommendations for learners across Bangladesh. Built, delivered, supported.
Voice2Policy
Citizen opinion classified at scale with a fine-tuned deBERTa model on AWS SageMaker, synthesized into whitepapers and published poll results.
Pulselinks ERP
Seven years leading a 29-engineer team delivering ERP across Switzerland and Bangladesh. Deployment time cut by 60 days. Outbound sales velocity up 330 percent.
Pre-Kitalon. Included because scope matters.
Products
We run our own products on the same discipline. They keep our deployment instincts honest and our opinions tested on real users.
GovProcurementAI
Managed proposal responses for certified government contractors. Submit an RFP, receive a submission-ready draft with a compliance matrix inside 24 hours.
govprocurementai.com →Restaurant Phone AI
A phone agent that answers every call, takes the order, and syncs it to the kitchen. Built on the OpenAI Realtime API and Twilio.
restaurantphoneai.com →Aether
A type-safe language for LLM orchestration with compile-time verification. MIT licensed.
View on GitHub →Founder
I don't wait for the right tools. I build them.
Kitalon Labs is built by Sowad Al-Mughni. Thirteen years across engineering, product, and executive roles. Led 29 engineers on cross-border ERP delivery at IMpulse (BD) Ltd. Shipped EdTech to 10,000+ students across 70 institutions as Technical Director at REAPRA. First place, PCH International Hardware Hackathon, Toronto.
- Interim CEO · Uddom
- AI & Technology Consultant · LEAD Academy
- AIPMM Certified Product Manager
- Writing code since age 10
Bring us the workflow that’s costing you the most.
Give us 30 minutes and the messy part of your operation. We'll tell you what can be automated, what should stay human, and whether building it is actually worth your time.