ler8.tech
Booked for 2026 · 1 slot Q1 2027

We build production software. AI is the xler8tor.

Bring us a real problem and a real deadline.
Your MVP is live in three weeks.

PracticeSoftware, AI, ML
LocationIslamabad / Remote
FootprintUS, EU, MENA

What we do

05 practices

We build software. AI is how we build it faster, and how we build what could not be built before.

Vision

Reads images, video, and handwriting. Vision models and VLMs.

Language

Reads and writes text. LLMs, NLP, fine-tuning, and our own models.

IoT

Edge devices, sensors, gateways, and telemetry. Firmware to dashboard, and the code that makes the readings agree.

Models

We pick the model per problem. Local open-weight for privacy or latency. Cloud API for capability and scale.

Machine learning

Finds patterns in business data. Anomaly detection, fraud classification, and scores that pass an audit.

All of it ships as apps. Mobile, web, embedded, any platform. Built for the real conditions of production: dropped frames, flaky networks, shifting requirements.

Selected projects

07 in production
Every project below is in production and under NDA. The diagrams are live simulations: the real mechanism, with illustrative data. The screenshots show the client’s product, obscured.
Filter
07 / 07 shown

Business Operations Intelligence

Multi-site operations · enterprise · 2025

A statistical engine running across 1,500+ locations. Each night it reads every feed the operator already pays for. It flags each deviation from normal, names what drives it, and prices what can be recovered.

  1. IngestEvery feed the operator already pays for, rebuilt each night into one modeled warehouse.
  2. ClassifyEach deviation sorted by what drives it: structural leakage, local drift, an integrity event, or transient noise.
  3. ExplainLanguage models write each finding three ways: for the site, the region, and the C-suite.
  4. PriceEvery finding checked against the figures and priced in recoverable dollars.

ResultAbout $10,000 recovered per location each year.

StackML anomaly & fraud
Modeled data warehouse
Proprietary LLM analysis
Video reconciliation
ROI pre-calculation
Footprint1,500+ locations
Checked every night
Findings priced in $
Business Operations Intelligence — confidential client UI, obscured
Business Operations Intelligence — confidential client UI, obscured

In production · confidential client UI, obscured

Read the full case study

Behavioral Simulation Engine

Workforce training · L&D · 2024

A training platform that replaces compliance videos and role-play workshops with a simulated workplace. Learners hold open-ended conversations with AI coworkers. Each coworker has a personality, a style, and trust killers: the behaviors that make it lose trust.

  1. CastAI coworkers, each with a personality, a style, a hidden agenda, and its own trust killers.
  2. ConverseOpen-ended conversation by text or by voice. The characters stay in role.
  3. ReactMishandle a colleague and their cooperation visibly drops.
  4. GradeFour dimensions, tied to the client’s own framework. Each finding anchored to the moment that triggered it.

ResultTraining takes a sixth of the time. It costs a tenth as much.

StackLocal fine-tuned LLMs
Personality-driven NPCs
Multi-pass scoring
Framework-bound rubrics
Grading4 dimensions
Tied to client framework
Verbatim evidence per finding
Behavioral Simulation Engine — confidential client UI, obscured
Behavioral Simulation Engine — confidential client UI, obscured

In production · confidential client UI, obscured

Read the full case study

Handwritten Census to X12 Claim

Healthcare revenue cycle · medical billing · 2026

A billing platform that replaces manual transcription between hospital rounds and the insurance claim. Vision models read every encounter off the marked-up census page, handwriting included. Each field links back to the exact place on the page it came from.

  1. ReadVision models lift every encounter off the marked-up census page, handwriting included.
  2. EvidenceEach field carries a confidence score and a box drawn back to the exact pixel it came from.
  3. CheckEight checks run before anything is sent, then a full round trip against the 837P standard.
  4. SubmitApproved encounters merge into an X12 837P file. Denials are read, sorted and refiled in the same tool.

ResultA census page took a reviewer 10 minutes. It now takes 1. Claim rejections fell by 60%.

StackVLM extraction
Bounding-box evidence
X12 837P generation
Closed denial loop
Validation8 checks before sending
Live CMS dictionary
CARC/RARC refile loop
Handwritten Census to X12 Claim — confidential client UI, obscured
Handwritten Census to X12 Claim — confidential client UI, obscured

In production · confidential client UI, obscured

Read the full case study

National Tax Code vs Climate Commitments

Climate finance · fiscal policy · 2026

A diagnostic tool for finance ministries. It tests whether the national tax code supports the climate commitments the country has signed. It sorts each provision as complementary, countervailing, overlapping, or redundant.

  1. ReadModels read the fiscal law statute by statute and weigh each provision against the country’s targets.
  2. SortEvery provision sorted: complementary, countervailing, overlapping, or redundant.
  3. GradeOne coherence score and a letter grade, weighted by how much each finding matters.
  4. SimulatePolicy teams assemble a reform package and watch the grade move in real time.

ResultHundreds of provisions read in a few minutes. By hand, the same review takes months.

StackLocal fine-tuned LLMs
Semantic graph
Conflict detection
Scenario simulation
OutputComposite coherence score
Letter grade
Reform simulator
National Tax Code vs Climate Commitments — confidential client UI, obscured
National Tax Code vs Climate Commitments — confidential client UI, obscured

In production · confidential client UI, obscured

Read the full case study

Throughput Verification & Counted-Unit Audit

Heavy industry · manufacturing & logistics · multi-site · 2024

A real-time counting system on the conveyors, dispatch belts, and loading gates where each unit crossing the line is a unit of revenue. Vision, a motion sensor, and a heat sensor validate each count independently. When they disagree, the count is flagged as contested. Each count is written to a tamper-evident audit log.

  1. TrackEach unit spotted, tracked as one identity, and counted once as it crosses the line.
  2. ValidateA motion sensor and a heat sensor confirm every camera count at the same line.
  3. ContestAny disagreement is flagged as a contested count. The system never picks a winner silently.
  4. LockEvery count written to a tamper-evident log the instant it is produced.

ResultCounts are 99% accurate, at a fifth of the cost.

StackCoreML / YOLO on-device
Motion sensor validator
Heat sensor validator
Trajectory tracking
Tamper-evident audit log
Throughput3 independent validators
Tolerates blocked views
3 parties, one record
Throughput Verification & Counted-Unit Audit — confidential client UI, obscured

In production · confidential client UI, obscured

Read the full case study

County Cybersecurity Posture

Government · cybersecurity · 2026

A security posture dashboard for a U.S. county government. It turns dozens of metrics from the county’s security tools into one county-wide score, built from weighted pillars. Readers drill down five levels, from the county to a single finding.

  1. CollectReads from the security tools the county already runs.
  2. NormalizeEvery metric on the same 0-to-100 scale, graded against its own thresholds.
  3. WeighPillars weighted into one number that adds up exactly. An average you can audit, not a hidden formula.
  4. DrillFive levels deep: county, pillar, department, user or device, finding.

ResultA posture report used to take a few weeks to months. It is now live on the dashboard, always.

StackPosture scoring
Compliance mapping
Drill-down dashboards
Multi-regime compliance
ScaleDozens of departments
Tens of thousands of endpoints
Five-level drill-down
County Cybersecurity Posture — confidential client UI, obscured
County Cybersecurity Posture — confidential client UI, obscured

In production · confidential client UI, obscured

Read the full case study

Tamper-Proof Evidence Chain

Environmental justice · human rights tech · 2025

A permanent, tamper-evident chain of custody for evidence of environmental harm. Each file is encrypted on the device. The unencrypted original never leaves it. The keys are split across trustees, so no single party can decrypt, the operators included.

  1. EncryptEvery file is encrypted on the device before anything is uploaded.
  2. AnchorA unique fingerprint pins the encrypted file to permanent storage and a public timestamp.
  3. SplitThe keys needed to decrypt are split across five independent trustees.
  4. OpenAny three trustees must agree. No single party can open the evidence alone.
StackArweave anchoring
Ethereum timestamps
AES-256-GCM (Web Crypto)
Shamir-style escrow
Trust model3-of-5 trustee quorum
0 IPs logged
0 admin reads
Tamper-Proof Evidence Chain — confidential client UI, obscured

In production · confidential client UI, obscured

Read the full case study

Who builds this

Founder · Team · Reach

Founder-led, with senior engineers engaged per project, not a standing team. 18+ years of production software at enterprise scale.

Co-founder and CTO at Oscar AI (2021 – 2023): operations platform for quick-service restaurant (QSR) franchises. $5M raised across two rounds. 1,500+ locations live. Distributed team across six time zones at peak.

Decade as VP of Software Development at SPS (2013 – 2023): scaled the engineering practice from 1 engineer to 35+ across Pakistan and the US. Delivered dozens of enterprise applications across logistics, utilities, healthcare, and real estate for clients including Asplundh, OnTrac, IBM, CREyield, and AllMyHealth. Several are still in active development and support today. Some carry 35K+ daily active users. First place globally across 70 countries (2016) for an IoT + NLP safety platform. IBM Cloud Champion 2017. MIT Innovation Leadership.

Founder’s profile on LinkedIn

Years
18+
In production software
Daily enterprise users
35K+
SPS-built apps, still running
Locations live
1.5K
QSR sites · Oscar AI
Raised
$5M
Oscar AI · co-founder, CTO

How we work

05 steps · MVP in 3 weeks

No wasted time. Not yours. Not ours.

One-page brief

Send us a brief: the problem, the deadline, the constraints. One page is enough. No forms.

First meeting

You walk us through the brief on a call. We ask what we need to know. If we are not the right fit, we say so on that call.

NDA & 24-hour proposal

If we both want to go on, we sign an NDA. Within 24 hours we send a preliminary proposal: questions, ideas, and a first approach. All free of cost.

Proposal & sign

If you like the direction, we send the full proposal. You sign it. We start work the same day.

MVP live in 3 weeks

Production-ready, monitored, and in use by real users. That is the guarantee. Not a roadmap, not a demo. Live.

Start a conversation

24h response

Tell us the problem.
Tell us the deadline.

We will tell you if we can ship it.

We respond within 24 hours. If we are the right fit, we move immediately.