Reported outcome from financial reconciliation automation.
MUHAMMAD ATTAYAB ASHRAF × PYTHON DEVELOPER
I build backend
systems and AI
that work.
I'm a backend engineer based in Lahore. I turn APIs, data, and language models into software that is useful, maintainable, and ready for real users.
BASED IN LAHORE, PAKISTANPYTHON · BACKEND · AI/LLM
01 / WHAT CHANGEDTHE PRACTICAL SIDE
Less manual work.
More useful work.
The best measure of engineering is the work it removes or improves. A few outcomes from my recent projects are worth noting.
Reported outcome from automated ETL workflows.
OCR-extracted textbook content in a RAG pipeline.
Reported project outcomes are CV-based; implementation details are kept private where needed.
02 / HOW I WORKStart with the problem.
Start with the problem.
Then choose the stack.
I like clear APIs, understandable data, honest tradeoffs, and code that a teammate can maintain without a mystery tour.
01 — BACKEND
Design the
foundations.
Build APIs and services that are easy to reason about: explicit contracts, well-shaped data models, secure access, and boundaries that can evolve.
FastAPIDjangoFlaskPostgreSQLREST
API contracts
DOMAIN logic
DATA models
CLEAR SERVICE BOUNDARIES02 — INTELLIGENCE
Make AI
useful.
Connect retrieval, language models, tools, and structured outputs to real product workflows — with orchestration and fallbacks designed in.
RAGFAISSEmbeddingsLLM APIsAgents
LLM
RETRIEVEROUTEGENERATECONTEXT → REASONING → OUTPUT03 — DATA & FLOW
Move work
off the request.
Use queues, scheduled workers, validation, retries, and stateful workflows to keep long-running operations observable and recoverable.
CeleryRedisETLWebhooksFAISS
INGEST
→PROCESS
→PERSIST
↺ RETRY / RECOVER
ASYNC BY DESIGN04 — PRODUCTION
Ship with
confidence.
Containerize, test, secure configuration, document operational behavior, and make deployment a repeatable engineering process.
DockerAWSGitHub ActionsNginxCI/CD
BUILD
↓
TEST
↓
DEPLOY
REPEATABLE DELIVERY03 / SELECTED WORKRecent work,
Recent work,
not just demos.
SCROLL TO EXPLORE →SELECT A PROJECT ↓01 — 03
01 / DEVELOPER TOOLINGPYTHON · AST · LLM
01class DocumentationPipeline:02def analyze(source):03ast = parse(source)04context = retrieve(ast)05return agents.verify(context)
ANALYSIS COMPLETE ●
MULTI-AGENT DOCUMENTATIONDocAgent
↗DocAgent
with C/C++ support
AST-level code analysis integrated into a multi-agent documentation pipeline with Reader, Searcher, Writer, and Verifier roles.
PYTHONLIBCLANGAST PARSINGLLM APIs
02 / APPLIED AIPYTHON · RAG · SEARCH
CHAPTER 04
semantic chunk
VECTOR
SPACE
SPACE
QUERY → RETRIEVE → ANSWER
DOCUMENT QUESTION ANSWERINGTextbook RAG
↗Textbook RAG
— Chemistry
Document processing and semantic question-answering built around OCR extraction, embeddings, vector retrieval, and LLM generation.
FAISSSENTENCE TRANSFORMERSPYMUPDFTESSERACT
03 / GENERATIVE MEDIAPYTHON · FLASK · FFMPEG
01 / SCRIPTSCENE
PROMPT
PROMPT
02 / GENERATE
03 / RENDER
SCRIPT-TO-VIDEO WORKFLOWAuraFLUX
↗AuraFLUX
Video Studio
AI-assisted production pipeline for scene planning, narration, image generation, background processing, and MP4 rendering.
FLASKPOSTGRESQLLLM APISFFMPEG
ALSO ON GITHUB
Smaller experiments, client-adjacent prototypes, and tools I've built along the way.
04 / CLIENT WORKProjects shipped.
Projects shipped.
Details kept discreet.
Some of this work belongs to clients, so I keep the code and sensitive details private. The role and engineering approach are still worth sharing.
ReconFY FINANCIAL RECONCILIATION SAAS
Automated matching across PDFs, spreadsheets, and POS data with a Python/FastAPI microservices architecture deployed on AWS.
~80%LESS MANUAL WORK
CV-REPORTED
CV-REPORTED
AI Tenant Screener VOICE AI PLATFORM
Telephony, speech recognition, configurable LLM providers, text-to-speech, and background task orchestration for applicant screening.
REAL-TIMEVOICE WORKFLOWS
ReadyRentalsOnline PROPERTY OPERATIONS
Work-order APIs, access control, audit logging, technician workflows, PDF reporting, and containerized VPS deployment.
END-TO-ENDWORK ORDER FLOW
ETL Automation DATA ENGINEERING
Validation, deduplication, and scheduled data processing pipelines built to remove repetitive manual operations.
~90%LESS MANUAL WORK
CV-REPORTED
CV-REPORTED
Client outcomes are CV-reported and should be supported with approved measurement context before being treated as independently verified benchmarks.
05 / TOOLKITTools I use
Tools I use
often and comfortably.
The stack follows the problem. These are the technologies I reach for most when building reliable products.
Backend
Python · FastAPI · Django · Flask · REST APIs · SQLAlchemy · Microservices
AI / LLM
RAG · FAISS · Embeddings · OpenAI · Claude · Groq · OpenRouter · Multi-agent systems
Data
PostgreSQL · MongoDB · MySQL · Redis · Pandas · NumPy · OCR · ETL
Infrastructure
Docker · Docker Compose · AWS EC2/S3/Cognito · Nginx · GitHub Actions · Linux
06 / A LITTLE ABOUT MEEngineer first.
Engineer first.
Builder by habit.
I studied Software Engineering at the University of Management and Technology in Lahore. Since 2024, I've worked across freelance product development and client engineering; today, I build backend systems at DevVerx.
I care about the whole path: understanding the problem, shaping the architecture, writing the code, checking it, and making it straightforward to run.
More about my experience ↗JUL 2026 — PRESENT
Python Developer DEVVERX
Backend services, APIs, integrations, and LLM-powered product features.
JAN 2024 — PRESENT
Freelance Engineer INDEPENDENT
End-to-end client projects spanning backend systems, AI workflows, and deployment.
2025
B.Sc. Software Engineering UMT · LAHORE
University of Management and Technology.
07 / SAY HELLOWorking on something
Working on something
worth building?
I’m open to backend and AI engineering roles, thoughtful collaborations, and conversations about useful software. A clear message is all it takes to start.