Product experience
Built substantial React and TypeScript feature slices across projects, meetings, results, evidence, stakeholder review and responsive product layouts.
02 · Overview · Requra.AI · Requirements engineering platform
From scattered context to review-ready requirements.
Requra.AI turns documents, meeting transcripts and project notes into structured requirements, user stories, acceptance criteria and source evidence that product teams can review, refine and deliver.
Generated requirements and stories remain inspectable and editable rather than arriving as one block of prose.
- Role
- Software EngineerContributed to the React product architecture, UI/UX, typed asynchronous workflows, AI job lifecycle, retrieval, grounding and quality controls, while aligning contracts across frontend, backend and AI services.
- Status
- In development · ITI graduation project
Screens use approved synthetic project data. No production customer data is shown.
02System story
Turn source material into work people can verify.
Project knowledge is scattered across documents, meetings and informal notes. The challenge is not generating more prose; it is producing structured work that keeps its source, status and review decision visible.
Synthetic source · Member onboarding
- 01
Source capture
Context enters with identity intact.
A document, meeting transcript or project note enters one workspace with its source retained.
- Status
- Transcript · accepted
- Evidence
- Member onboarding · line 118
- 02
Evidence grounding
Claims remain tied to source.
Retrieval finds supporting fragments, then grounding verifies them against the original source.
- Status
- 3 fragments · source checked
- Evidence
- Hybrid retrieval
- 03
AI processing
Durable work exposes its state.
Queued, processing, partial, failed and retryable states stay visible while structured analysis runs.
- Status
- Queued → processing → partial
- Evidence
- Attempt 01 · recoverable
- 04
Structured candidate
Prose becomes reviewable structure.
The result arrives as a typed requirement candidate with priority, evidence and decision state.
- Status
- REQ-SYN-014 · candidate
- Evidence
- Functional · high priority
- 05
Human review
A person decides what becomes truth.
A reviewer edits, approves or rejects the candidate; raw AI output cannot cross this gate alone.
- Status
- Decision required
- Evidence
- Edit · approve · reject
- 06
Traceability
Approved work keeps its source.
The resolved requirement remains linked to the evidence and human decision that made it deliverable.
- Status
- Approved · source linked
- Evidence
- Ready for delivery
03What I built
What I contributed across the system
AI workflow
Contributed to durable FastAPI jobs, Redis/RQ worker execution, retrieval, evidence grounding and quality controls.
Cross-service alignment
Helped align typed contracts across the React frontend, ASP.NET Core backend and FastAPI pipeline.
The ASP.NET Core implementation was team-owned and is not presented as my work.
04Engineering decisions
Making AI output usable and accountable
Long-running analysis is a job, not a request
- Problem
- Document analysis can take longer than a normal request and may partially fail.
- Decision
- Persist analysis as durable background work with explicit queued, processing, partial, failed, retryable and completed states.
- Result
- Users can understand progress, recover from failure and return to results without relying on one loading screen.
Retrieval is not proof
- Problem
- Relevant source text can still be incorrectly attached to generated output.
- Decision
- Retrieve supporting context, then separately verify that the final quoted evidence exists in the original source chunks.
- Result
- Reviewers can inspect where a requirement came from instead of trusting model output blindly.
Generated output remains a candidate
- Problem
- AI-generated requirements may appear complete before a person has validated them.
- Decision
- Keep generated output in an editable candidate state with approve, edit and reject paths.
- Result
- Only reviewed work becomes an approved requirement.
System path
- React · TanStack Query · Zustand
- ASP.NET Core
- FastAPI · Redis/RQ
- LangGraph
- PostgreSQL · pgvector
05The product in use
The product in use
01 / 02
Project workspace
Documents, meetings and analysis runs remain connected to one project context.
02 / 02
Source traceability
Each requirement keeps a visible path back to the project context that supports it.
Screens use approved fictional or synthetic data. No production user data is shown.
Requra turns fragmented project context into structured work that remains traceable, reviewable and ready for delivery.