Human-Approved AI Agent for Filing and Tracking Citizen Service Requests
Filing a complaint or service request with a city or government office means finding the right portal, filling long forms, attaching documents and then chasing it for weeks. Build an agent that turns a citizen's plain description (in Urdu, Roman Urdu or English) into a correctly filled request, pauses for the citizen to approve every submission, then tracks status and reminds them to follow up. Built and tested against a realistic mock portal you create.
Advanced
Needs a team, real planning, and a supervisor who knows the area.2 semesters · 2 to 3 students
7 to ship
3 optional extrasSuggested stack
What you should ship
- A mock government service portal you build, modelled on common patterns: categories, multi-step forms, attachments, CAPTCHA placeholder and a status page
- An intake conversation that understands a problem described in Urdu, Roman Urdu or English and picks the right category and department
- A form-filling step (structured API where available, browser automation with Playwright on the mock portal) that never submits on its own
- An approval screen that shows every field and attachment exactly as it will be submitted, with edit and cancel
- A tracker that checks status on a schedule, summarises changes and reminds the citizen when a deadline passes
- Guardrails: no storage of passwords, consent recorded per request, and a full log of agent actions
- An evaluation on 60 scripted citizen scenarios: correct category, correct fields, approval respected, and time saved
If you have time left
- Voice notes as input
- A help-desk mode for NGO or union council volunteers who file on behalf of people who cannot use a phone
- Templates for writing a polite follow-up letter when a request stalls
The problem: government services are a maze
A broken streetlight, a delayed pension, a wrong electricity bill. Pakistan's Citizen Portal, launched in 2018, and similar portals in other countries made it possible to file complaints online, and millions have. But many people still struggle: which department, which category, what to attach, and how to follow up when nothing happens. People with low literacy or little time give up.
Browser agents can now fill forms and click through websites. That is useful here, and also risky: an agent that submits the wrong request, in the wrong name, or without the citizen's knowledge does real harm.
What already exists and the gap
- Government portals and apps provide filing and tracking, but expect the citizen to know the system.
- Browser agents such as OpenAI's Operator-style agents and open source projects like browser-use can operate websites, with varying reliability.
- Volunteer help desks run by NGOs help people file in person.
The gap is an agent designed around consent: it does the tedious work, but every submission is approved by the citizen after a clear preview, and it keeps chasing until the request is closed.
Inspiration
Anthropic's guidance on building effective agents, and the OWASP Top 10 for Agentic Applications, both stress human oversight for consequential actions. Your project applies that to a public service where trust matters more than speed.
What you build
- A mock portal that behaves like a real one, so you can test safely.
- Intake, form filling and approval as separate, testable steps.
- A tracker with reminders.
The technically interesting part
- Mapping a messy description to a category. "Gali mein gutter ubal raha hai" should become a sewerage complaint for the right area.
- Robust form filling. Forms change, fields are optional or conditional, uploads fail. Handle each failure visibly.
- Approval that means something. The preview must be precise and readable. Measure whether testers catch planted mistakes.
How you will evaluate it
- Accuracy of category and field filling on 60 scenarios.
- Planted-error catch rate in the approval step.
- Time to file with and without the agent, with a few volunteer testers.
Path to a product
NGOs, legal aid clinics and community help desks could use an assisted-filing tool, and governments themselves may want a conversational front door to their portals. The safe-by-design approval pattern is reusable for any agent that acts on someone's behalf.
Scope warning
Do not run the agent against real government portals or file real requests during the project. Build and test on your mock portal. Never ask users for their portal passwords. Keep the number of request categories small.
Learn what you need on Sefism
- AI agents roadmap: planning, tools, memory and human-in-the-loop design.
- AI security roadmap: agent risks and why approval steps matter.
- Introduction to LangGraph: building agents with explicit steps and interrupts.
- Playwright documentation: browser automation for the form-filling step.
- OWASP Top 10 for Agentic Applications: the risks your guardrails address.
- How to prove an AI agent project works: structuring your evaluation.
Get help with this idea
The risk is an impressive demo that would be unsafe in real use. Tauseef can help you design the approval flow, the mock portal and an evaluation that tests safety as well as accuracy. If you are new here, create a free account first.
Ideas and guidance, not finished projects
These are project ideas and scoping guidance, published free for students to use as a starting point. I do not build, write, or sell final-year projects, and I do not complete coursework for anyone. Take an idea, make it yours, and build it.
Not sure this one fits you?
I help students pick an idea that matches their skills and their deadline, then scope it down until it is finishable. Guidance only, never done for you.
Stuck on something specific?
Writing only gets you so far. If you want an answer to your situation rather than the general case, book a session and we will work through it together. Sessions are free for approved Sefism members, and a few slots open each week.
Follow along
New writing, resources and project ideas land here first.