Technology Stack
Next.js, TypeScript, Tailwind CSS, shadcn/ui, React, Vite, Chrome Extension APIs, OpenAI, WebRTC, IndexedDB, AWS Polly, Vercel
Overview
Duration: 12/2024 - ongoing
Let's Talk Wise helps candidates prepare for job interviews by actually putting them through one. The platform runs an AI interviewer that asks role-appropriate questions, listens to the answer, and returns structured feedback on both what was said and how it was said — the two things a candidate cannot assess about themselves in the moment.
It ships as two surfaces: a web application for full practice sessions and progress tracking, and a Chrome extension that lets a candidate start a tailored mock interview from the job listing they are looking at, at the moment their motivation is highest.
Interview Engine
- Adaptive questioning — the next question is chosen from the previous answer, so a session follows the shape of a real interview rather than reading down a fixed list
- Role and industry targeting — question sets specialised by profession and seniority, so a backend engineer and an account manager do not get the same interview
- Spoken interviewer — questions delivered as natural speech through AWS Polly, so the candidate practises listening and responding under time pressure rather than reading
- Answer evaluation — responses assessed for relevance, structure, and persuasiveness against the question that was asked
- Delivery analysis — pacing, filler words, and tone measured from the audio, which is where most interview coaching actually lives
- Multi-language practice for candidates interviewing outside their first language
Chrome Extension
The extension is the distribution strategy as much as a feature. It activates on job listings across major career sites, reads the role context from the page, and launches a practice session tailored to that specific posting — turning a passive browsing moment into a practice session without the candidate ever navigating to the product.
Technical Architecture
- Real-time media — WebRTC for low-latency audio and video capture, so analysis happens against the live session rather than an uploaded file after the fact
- Speech synthesis through AWS Polly for the interviewer voice
- Extension architecture — content scripts reading listing context, a background service worker holding session state, and a shared authentication bridge with the web application
- Local-first storage — IndexedDB holds practice material and session data on device, keeping recordings out of transit where they do not need to be there
- Shared component layer — the web app and the extension UI draw on the same design system, so the two surfaces stay visually and behaviourally consistent
- Edge-deployed API routes for the latency-sensitive parts of the loop
Candidate Experience
- Session dashboard — practice history and improvement over time, so a candidate can see progress rather than guessing at it
- Timestamped feedback — comments anchored to the moment in the recording they refer to, which is the difference between actionable and abstract
- Custom sessions — practice targeted at a specific role or company
- Progress tracking across repeated sessions on the same question set
- Accessible by construction — full keyboard navigation and screen reader support
- Light and dark themes
Privacy
Interview recordings are personal data of an unusually sensitive kind — someone's worst answers, recorded. The platform treats that accordingly:
- On-device processing wherever the analysis can be run locally, minimising what leaves the machine at all
- Encrypted storage for sessions that do need to be retained
- User-controlled retention — export and deletion available without a support request
- Transparent scoring — the platform explains what it evaluated and why, rather than returning an unexplained number
This project combines a real-time media pipeline, a browser extension with its own distribution logic, and an AI evaluation loop, into a product where the technical difficulty is invisible and the value is immediate.