FIRST PLACE
MIA Trek Pitch Competition
Countify · AI for scientific discovery
Engineer by craft. Founder by instinct.
OPEN TO OPPORTUNITIES WORLDWIDEFrom model
potential to
production impact.
I’m Faisal, an AI engineer and technical founder. I turn ambiguous customer problems into working systems, then build the evaluation that earns their trust.
Faisal RehmanAI TECH LEAD @ CHECKMATE
Building reliable AI, end to end.
RECOGNITION
Award-winning work.
WINNER
IDEA Lab Venture Competition
Conversational voice ordering startup
NATIONAL DATA SCIENCE AWARD
NESCON ’19
Team representing CareCloud
HONORABLE MENTION
Pakistan Harvard Conference 2026
Countify · Pitch competition
01 SELECTED WORK
Built, shipped, measured.
The systems behind the numbers. My ownership, the engineering decisions, and what changed.
VoiceBite / Task completion
Better systems. Measurable progress.
Evaluation → failure attribution → targeted fixes
Making a voice agent
reliable enough to order.
Live conversations expose every weak link. I led core engineering for VoiceBite and built the evaluation function to separate model failures from speech, infrastructure, and data failures.
Low 70s → 90%+task completion in a single-location pilot
Inside the system “Help me understand and act on my data.”
Request→MCP tools→Action
Answers, insights, and actions in Checkmate.
From business questions
to actions inside Checkmate.
An AI assistant that answers business questions, generates insights, and uses MCP to take actions within Checkmate. I built the product end to end.
MCP-poweredinsights and actions inside the product
Read the case study RESEARCHER → MCP01 / Plan the pipeline02 / Segment & count03 / Validate & reproduce
A research question.
A reproducible vision pipeline.
MCP-orchestrated agents turn natural language into scientific image analysis, with specialized vision tools and a researcher validating each step.
1st placeMIA Trek Pitch Competition
Read the case study ZERO → ONE2025
A product people paid for.
Paying locations12
Pilot → paid80%
Customer retention100%
From customer discovery
to twelve paying locations.
I took a voice ordering product through discovery, prototyping, founder-led sales, pilots, deployment, and support. The job was to make the technology work for restaurant owners.
80%pilot-to-paid conversion · 100% retention
From zero to one 3D symptom captureIllness summaryDiagnosis → CPTMedication & lab suggestions
The patient encounter,
connected end to end.
A 3D symptom interface that connects patient intake to an illness summary, diagnosis and CPT codes, and medication and lab-order suggestions.
End to endpatient–provider encounter support
Explore the project Rich vehicle routing · conceptual illustration
Rich vehicle routing.
Real operating constraints.
Built SHARE’s proprietary routing solution, balancing vehicle capacity, time windows, and operating cost to serve real mobility demand.
40% lowerrouting cost · 22% fewer fleet miles
Explore the project OperationsVehicle telemetryBusiness systems
Company-wide data warehouse
5 internal teams60+ client companies
The data foundation
behind a mobility business.
Built the company-wide warehouse and analytics platform from scratch, bringing operational data into shared reporting for internal teams and client companies.
99.9%daily pipeline reliability · 40+ dashboards
Explore the project Commute planning that became a business.
SHARE Mobility · A commute simulation and fleet-planning platform that became standalone SaaS and helped support a $12M raise.
Explore commuter analysis ↗Clinical documents into usable data.
CareCloud · Document intelligence processing 60K+ documents daily, with claim cycles reduced from 24 hours to 6.
Explore document intelligence ↗02 ENGINEERING JUDGMENTIntelligence is a capability.
Intelligence is a capability.
Reliability is a discipline.
My work lives at the boundary between what a model can do and what a customer can depend on.
Put guarantees in code.
Models interpret intent. Deterministic code enforces permissions, validates actions, and decides what gets submitted.
Fix the layer that actually failed.
A suspected model failure in VoiceBite traced back to transcription. Targeted changes cut that error class about 50%.
Make evaluation part of the release.
Production failures become regression cases. Simulation suites test the full workflow. LLM graders are checked against human labels.
Own the outcome beyond the model.
Customer discovery, interface, latency, deployment, and support belong in the same engineering conversation.
03 BUILDING IN PUBLICCode you can inspect.
Code you can inspect.
Thinking you can read.
Small, concrete examples of how I approach retrieval, model adaptation, and evaluation.
RxNorm Voice Normalizer
A small agent that maps noisy medication names to canonical RxNorm entries. The LLM proposes candidates; every identifier must validate against the source of truth.
The design, in three steps01 Try a direct RxNorm match
02 Propose candidates if needed
03 Validate before returning an ID
02 Propose candidates if needed
03 Validate before returning an ID
15 curated evaluation cases · Category-level accuracy · p50/p95 latency · Prompt experiments
Explore the repositorySEP 2025 · 2 MIN READ · RETRIEVAL
All writing on Medium ↗When customers say “caafee,” they want coffee.
Combining text similarity, semantic search, and phonetic normalization to resolve noisy voice inputs.
FEB 2025 · 2 MIN READ · FINE-TUNINGFine-tuning large models with limited resources.
A practical introduction to low-rank adaptation and why smaller trainable updates can be useful.
04 IN THEIR WORDS
What colleagues say.
Perspectives from people I’ve built with, reported to, and mentored. Selected excerpts from their LinkedIn recommendations.
“He knows what’s worth measuring, what actually matters to the product, and how to explain complicated technical problems in a way that makes sense to everyone in the room.”
Christopher Lam Senior colleague · Checkmate
“He introduced structured evaluation frameworks, prompt versioning, regression testing, failure analysis, and data-driven experimentation.”
Jeffrey Gordon Direct manager · Checkmate
“Plenty of people can write prompts. Very few understand how model behavior, evaluation design, data quality, observability, analytics, engineering systems, and business objectives all interact. Faisal does.”
Jeffrey Gordon Direct manager · Checkmate
“I also got to watch him grow as a manager while I was at Share. He is a great mentor for younger devs who want to learn about ML / Data Pipelines / Complex Algs.”
Alexander Bothe Senior colleague · SHARE Mobility
More from colleagues 3 more recommendations
“His knowledge of Data Science and Data Engineering allows him to walk into complex problems and handle them gracefully.”
Christien Kelly Teammate · CEO at Trackore
“Faisal has a thorough understanding of statistical models, data visualization methods, and machine learning algorithms, which enables him to precisely tackle challenging issues.”
Suleman Shehzad Teammate · GenAI Lead
“Independent and very kind-hearted and supportive person. Innovative perfectionist always ready to put all his energy and stamina to get the job done.”
Umair Shah Direct report · AI Engineer
Want the full picture? Read every recommendation, word for word, on LinkedIn.
Read all recommendations on LinkedIn05 THE THROUGHLINEA career in
A career in
making it work.
From data systems and clinical NLP to optimization and agentic products. A consistent focus on the real workflow around the technology.
→ PRESENT
Checkmate
Senior AI & LLM Engineer / AI Tech Lead
VoiceBite: core engineering and evaluation for a production pilot. Ask Checkmate: an AI product assistant for insights and MCP-powered actions within Checkmate.
→ OCT 2025
Conversational Voice Ordering Platform
Co-Founder · AI startup
Built the product and the early business: customer discovery, founder-led sales, 12 paying locations, and deployment ownership.
2024–2025
SHARE Mobility
Data Science Lead, AI & Optimization Pod (prior: Data Scientist)
Built SHARE’s proprietary rich vehicle routing solution, company-wide data warehouse, and commuter analysis platform. Led cross-functional engineering and partnered with 100+ enterprise clients and public agencies.
→ APR 2021
CareCloud Inc. (formerly MTBC)
Data Scientist / Data Science Lead (prior: Jr. Data Scientist)
Built an end-to-end EHR encounter assistant, from a 3D symptom interface to illness summaries, diagnosis and CPT codes, medication and lab suggestions. Also built document intelligence and claim-denial models.
→ SEP 2019
Finalytica
BI Developer · Part-time
Built reporting and data models with Teradata, SQL Server, and Power BI.
06 A LITTLE MORE CONTEXTAn engineer with
An engineer with
a founder’s perspective.
I like taking a problem from the first customer conversation all the way to a system people use. That has meant selling pilots, designing data pipelines, debugging real-time agents, and building tools for researchers.
I’m currently building at Checkmate and developing Countify, an AI platform for scientific image analysis.
M.S. Data Analytics EngineeringNortheastern University · Boston · 2024
B.S. Software EngineeringInternational Islamic University, Islamabad · 2019
AWS Certified Solutions ArchitectAmazon Web Services
Agents & retrieval
Anthropic (Claude) and OpenAI APIs · Tool use · MCP · Pydantic AI · LangGraph · Hybrid RAG · pgvector · Reranking
Evaluation & model adaptation
Simulation suites · Human-calibrated LLM graders · Failure attribution · Langfuse · Coval · QLoRA / LoRA · PyTorch
Product engineering
Python · TypeScript · FastAPI · React / Next.js · WebSockets · Real-time voice · PostgreSQL · Redis
Data & infrastructure
AWS · Snowflake · DuckDB · Spark · Docker · Kubernetes · Terraform · CI/CD · Optimization
07 WHAT’S NEXTHard problems.
Hard problems.
Meaningful work.
I’m interested in teams bringing capable AI into the real world. Let’s talk about AI engineering, agent reliability, and products worth building.
faisalkhattak391@gmail.comOpen to opportunities in any location