I’m Wasim Ranjha. I build web applications, AI agents, and automation that connect your tools and take repetitive work off your team.
Work directly with me. From the first workflow to deployment.
Python
Django
FastAPI
React
PostgreSQL
AWS
Qdrant
PlaywrightVoice AI, business platforms, and browser automation. A selection of the products I’ve built, with interfaces and capabilities you can explore.
Showing 3 of 23 projects
One onboarding flow connects a business account, a dedicated number, and an AI assistant.
Connecting account setup and telephony into one usable product, so the business can configure its assistant around its own phone number.
Equipment records, maintenance schedules, and fleet reports share one operational workspace.
Keeping the asset and its maintenance information together, instead of treating scheduling and reporting as disconnected tasks.
A visual builder turns repeated browser steps into a workflow that can be used again.
Making a sequence of browser actions understandable as a workflow, so the user can see how the repeated task is structured.
A tool that breaks a goal into tasks and coordinates AI agents to work through them.
A web application for reviewing property listings and generating analysis reports.
An Android app for reading and listening to the Quran, checking prayer times, and finding the Qibla.
An AI proposal tool and Chrome extension built around Gemini and retrieval from reference material.
A dealership application for managing vehicles, sales, and customer information.
A hiring application with resume screening, candidate ranking, and interview scheduling.
A healthcare platform for patient records, billing, prescriptions, and day-to-day clinic operations.
A document-signing tool with support for multiple signers and audit trails.
A point-of-sale application for restaurant orders, kitchen workflows, and billing.
A retail point-of-sale application for sales, receipts, and stock tracking.
A medical-store point-of-sale application for prescriptions, stock, and billing.
A dental practice application for appointments, patient records, treatment plans, and billing.
An agency tool for managing projects, teams, and invoices.
An administrative platform for operations, HR, and compliance tracking.
A Shopify monitoring tool that tracks product prices and stock changes.
Content scheduling and automation across social media platforms.
A product-data collection tool for Temu, including prices and exports.
A tool for collecting product details, prices, and reviews from AliExpress.
A data collection tool for vehicle listings on PakWheels.
A conversational assistant using retrieval, LangChain, and OpenAI to work with context.
My core work is AI, web applications, and automation. For mobile and creative projects, I bring in specialist collaborators.
Not sure where to start?Let’s talkPhone agents, chat assistants, and search tools connected to your documents and business software.
Python / FastAPI / RAGView scopeClose scopeCustomer portals, internal tools, and SaaS products. Frontend, backend, and deployment handled together.
React / TypeScript / PythonView scopeClose scopeCollect and clean the information you need, then deliver it to your database or team.
Python / Playwright / SeleniumView scopeClose scopeSet up the infrastructure, databases, and deployment process around your application.
AWS / DigitalOcean / LinuxView scopeClose scopeiOS and Android development with my specialist team, supported by the APIs I build.
Flutter / React Native / Mobile UIView scopeClose scopeInterface design, brand assets, and video production with my creative team.
UI design / Brand identity / GraphicsView scopeClose scopeWasim Ranjha · Full Stack AI Engineer
Also known as Waseem Ranjha
I’m Wasim, a Full Stack AI Engineer working across the interface, backend, and AI layer. I build with React, TypeScript, Python, Django, FastAPI, and Flask.
You work directly with me on the technical build: defining the workflow, developing the application, connecting your tools, and preparing it for deployment.
I built Kira AI, a voice receptionist that answers inbound calls and books appointments for clinics and real estate businesses.
SaaS platforms, voice agents, FastAPI backends, and AWS deployments.
Django APIs and React features for applications across multiple industries.
Hands-on application development with Django and Flask.
ICS — Computer Science, Degree College Kot Momin (2018–2020). Professional training and mentorship in Python and backend systems at NeXskill.
We agree on the scope first, then work in small, reviewable steps.
We map your users, current tools, and the task that needs to work better.
We agree on the first release, architecture, milestones, and estimate.
You see working versions, give feedback, and help shape the next step.
We test the agreed workflows, deploy, and document how to operate the application.
Illustrative project briefs showing how AI, web applications, and automation can fit into everyday business workflows.
Incoming questions and appointment requests interrupt work throughout the day.
An AI receptionist handles agreed questions and booking requests, with a clear path to your team.
Customer records, tasks, and updates live in disconnected files and messages.
A shared web application brings the core workflow into one place with the right access for each person.
Someone checks the same sources and copies information into a spreadsheet every week.
A scheduled pipeline collects agreed fields, validates the data, and delivers a structured export.
Start with the problem you want to solve. I’ll propose the scope, milestones, and estimate before we begin.
For a repeated task that needs a better workflow.
For a new product or the next stage of an existing one.
For voice agents and AI connected to your business.
These are hourly development rates. Your proposal confirms the estimated hours, deliverables, support duration, and included support work. Hosting, AI usage, and other third-party costs are identified separately.
Practical notes on retrieval, voice agents, browser automation, and the choices that make systems maintainable.
A useful document search tool needs more than an upload button. It needs to keep track of where each passage came from, who may read it, and whether the source has changed. I would settle those requirements before choosing a model.
A practical starting point is to split documents into passages, store their embeddings alongside source references, and retrieve relevant passages for each question. pgvector adds vector similarity search to PostgreSQL, so the content and its metadata can live together.
Show the source passages with the answer and let the user open the original document. Test with real questions, including questions the documents cannot answer. The useful result is an answer someone can check—not simply fluent text.
I would begin by writing down the job in ordinary steps: collect information, check it, prepare a result, and request approval if needed. If a single workflow can do that reliably, adding several agents may only make it harder to debug.
For a multi-agent design, define the input and expected output of each step. Keep shared state explicit, record failures, and put limits on retries and execution time. A task should not continue indefinitely because two agents disagree.
Drafting a response and sending it are different actions. Separate suggestions from changes to external systems, and decide which actions need human review. Start testing with recorded examples before allowing the system to operate on live work.
The starting point is the team’s language, browser requirements, and existing automation. A working Selenium suite can be worth extending. For a new project, I would compare a small end-to-end flow in each candidate rather than choose from a feature list alone.
Playwright’s locators support finding elements by role, label, and other user-facing properties. These are useful when the goal is to interact with the application as a person would. Selectors based only on a long chain of page containers tend to be harder to maintain.
Authentication, slow pages, and layout changes are part of the work. Keep test data separate, capture useful failure information, and make retry behavior deliberate. A script that succeeds once is the beginning, not the finished automation.
An assistant might be missing information, using the wrong format, or making the wrong decision. Those are different problems. I would collect representative examples and define what a good response looks like before changing the system.
For missing business information, a retrieval step may be worth evaluating. For an inconsistent format, clearer instructions and output validation may be enough. Fine-tuning is a candidate to investigate when a repeatable behavior remains difficult to achieve and suitable training examples are available.
Keep a separate evaluation set and compare quality, cost, and response time. Include difficult and ambiguous cases. The decision should come from those results, with a clear plan for maintaining the chosen approach as requirements change.
Kira supports medical and dental clinics, real estate, plumbing, HVAC, and other business niches. Businesses register an account, get a phone number, and connect an AI assistant to handle incoming calls using Twilio and Vapi. The wider design question is where an automated receptionist should help and where a staff member should take over.
For an appointment workflow, I would define how availability is checked, how details are confirmed, and what happens when a booking system is unavailable. The agent should not say an appointment is booked until the connected system confirms it.
A useful handoff gives staff the caller’s request and the information already collected, within the agreed privacy requirements. Test interrupted speech, corrections, and unsuccessful calls as carefully as the happy path.
List the sources, fields, update frequency, and delivery format. Decide how to represent missing values and duplicate records. These choices determine whether the collected data is useful to the team receiving it.
Request pacing should account for the source and its access requirements. Scrapy’s AutoThrottle adjusts download delays using latency and a target concurrency setting, within configured limits. Whatever framework is used, errors should lead to bounded retries rather than an unlimited request loop.
A completed job can still contain incomplete or unexpected data. Validate required fields, monitor record counts, and keep enough source context to investigate changes. Save progress so a failure does not require starting every collection again.
AI voice and chat agents, RAG systems, SaaS products, internal tools, scraping pipelines, and cloud infrastructure. My specialist team also supports mobile development and creative work.
Yes. We start by reviewing your current software and available APIs, then scope the connections your workflow needs. Integration requirements are agreed before development.
Hourly development rates are $15 for focused automation, $19 for web apps, and $49 for advanced AI systems. Your proposal confirms the estimated hours, scope, milestones, and support. Hosting and third-party usage costs are identified separately.
We agree on handover and support before development starts. Your proposal specifies the support duration, included work, and response expectations, so you know how maintenance and future changes will be handled.
Absolutely. Bring one process, one product idea, or one technical problem. We can define a focused first release and build on it as your needs become clearer.
Bring the idea, the messy workflow, or the product that needs its next chapter. We’ll work out a useful first step together.
[email protected] +92 343 6000542Prefer WhatsApp? Let’s chat