Skip to content
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
36 commits
Select commit Hold shift + click to select a range
96f1d8e
docs: spec + implementation plan for Ukrainian language support
danilarff86 Sep 1, 2026
c12f242
feat: Ukrainian answer-language rule in prompt builder
danilarff86 Sep 1, 2026
d10018d
test: knowledge-base chunking is Cyrillic-safe
danilarff86 Sep 1, 2026
9998e0f
feat: route Ukrainian STT to Deepgram nova-3
danilarff86 Sep 1, 2026
57e06ba
feat: Ukrainian option in transcript- and answer-language selects
danilarff86 Sep 1, 2026
4ca8c8c
docs: changelog entry for Ukrainian language support
danilarff86 Sep 1, 2026
4d724c3
style: prettier formatting for Cyrillic chunking test
danilarff86 Sep 1, 2026
87acf6f
Merge pull request #1 from danilarff86/ukrainian-language-support
danilarff86 Sep 1, 2026
1ce636a
docs: design and implementation plan for academic answer style
danilarff86 Sep 1, 2026
076cd9d
feat: academic prose answers grounded in dissertation materials
danilarff86 Sep 1, 2026
7ae6314
feat: extract committee questions instead of interview questions
danilarff86 Sep 1, 2026
4983bcd
feat: raise answer length budget to 1200 chars with settings migration
danilarff86 Sep 1, 2026
d72a645
docs: describe academic prose answers and the new length default
danilarff86 Sep 1, 2026
cb2c67c
fix: drop interview framing from the extraction cue
danilarff86 Sep 1, 2026
90c08db
Merge pull request #2 from danilarff86/academic-answer-style
danilarff86 Sep 1, 2026
7d8fcc5
docs: add spec and plan for macOS system-audio loopback fix
danilarff86 Sep 8, 2026
683ede1
docs: add spec and plan for the Electron 44 upgrade
danilarff86 Sep 8, 2026
1a7b5c4
docs: add comment-translation task to the Electron 44 plan
danilarff86 Sep 8, 2026
5f09ee5
refactor: translate source comments to English
danilarff86 Sep 8, 2026
a7363c5
feat: declare NSAudioCaptureUsageDescription on both packaging paths
danilarff86 Sep 8, 2026
731079c
fix: drop the inert loopback feature-flag module
danilarff86 Sep 8, 2026
9a9bd5d
docs: fix the Electron 44 capability check to search the implementation
danilarff86 Sep 8, 2026
544f242
chore: upgrade Electron 33 -> 44 for macOS CoreAudio Tap audio capture
danilarff86 Sep 8, 2026
6b43486
docs: describe the Electron 44 system-audio upgrade
danilarff86 Sep 8, 2026
4f30248
fix: close the Node-floor and stale-reference gaps found in review
danilarff86 Sep 8, 2026
5ce40ca
Merge pull request #3 from danilarff86/fix/macos-system-audio-electro…
danilarff86 Sep 8, 2026
7a776a1
fix: translate remaining Chinese error/UI strings to English (#4)
danilarff86 Sep 8, 2026
dd8f053
fix: commit a transcript turn on end of speech, not on every frozen c…
danilarff86 Sep 8, 2026
721c0d5
fix: make the extracted "Current Question" self-contained (#6)
danilarff86 Sep 8, 2026
f32119a
docs: add model picker + Anthropic implementation plan
danilarff86 Sep 25, 2026
b96b8e5
feat: add Anthropic Messages API streaming client
danilarff86 Sep 25, 2026
8813d76
feat: register Anthropic provider; move resolveProvider into config
danilarff86 Sep 25, 2026
0832073
feat: list provider models live with a curated fallback
danilarff86 Sep 25, 2026
9a6d03e
feat: expose model listing and selection over IPC
danilarff86 Sep 25, 2026
94921ef
feat: model picker in the top bar; Anthropic key in Settings
danilarff86 Sep 25, 2026
bf12f05
fix: keep the top bar on-screen; give Claude's extraction call room t…
danilarff86 Sep 25, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions .env.example
Original file line number Diff line number Diff line change
Expand Up @@ -11,5 +11,6 @@ DEEPGRAM_API_KEY=
DEEPSEEK_API_KEY=
GEMINI_API_KEY=
OPENAI_API_KEY=
ANTHROPIC_API_KEY=
# Ollama runs locally and needs no key; default base URL:
# OLLAMA_BASE_URL=http://localhost:11434/v1/chat/completions
2 changes: 1 addition & 1 deletion .github/workflows/ci.yml
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@ jobs:
- uses: actions/checkout@v4
- uses: actions/setup-node@v4
with:
node-version: 20
node-version: 22
cache: npm
- run: npm ci
- run: npm run lint
Expand Down
2 changes: 1 addition & 1 deletion .github/workflows/release.yml
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,7 @@ jobs:
- uses: actions/checkout@v4
- uses: actions/setup-node@v4
with:
node-version: 20
node-version: 22
cache: npm
- run: npm ci
- name: Build & publish installers
Expand Down
2 changes: 1 addition & 1 deletion .nvmrc
Original file line number Diff line number Diff line change
@@ -1 +1 @@
20
22
8 changes: 8 additions & 0 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -6,12 +6,20 @@ All notable changes to this project are documented here. Format loosely follows
## [Unreleased]

### Added
- **Anthropic (Claude) provider and a model picker.** Choose the provider in Settings (now including Anthropic), and the model from the new **Model** dropdown in the top bar. The list comes live from the provider's API (OpenAI, Gemini, Anthropic, DeepSeek, Ollama) and falls back to a built-in list when there is no key or the provider can't be reached; ↻ refreshes it. Each provider remembers its own last model. The model text fields were removed from Settings.
- **Ukrainian language support**: `Українська` is now selectable as the transcript language (Deepgram `nova-3`, `language=uk`) and as the answer language (answers generated in Ukrainian). Ukrainian documents in the Knowledge Base were already supported and are now covered by tests.
- **Windows & Linux support for running**: npm scripts are now cross-platform (`cross-env`), so `npm start` works on Windows (previously the `VAR=` prefix broke cmd). `make run` adapts to the OS — macOS builds & opens the signed app; Windows/Linux launch via `npm start`. Prebuilt `.exe` (Windows) and `.AppImage` (Linux) ship in Releases.

### Fixed
- The first-run prompt to configure API keys now checks the selected provider's key instead of always asking for a Gemini key.
- **Sentences are no longer split at the slightest pause.** The transcript treated Deepgram's `is_final` flag as end-of-sentence, but that flag only means "this chunk won't be revised" — Deepgram emits several per spoken sentence. One question therefore arrived as two or three separate turns, and the fragments were then handed to the LLM, which stitched them into a garbled question. Frozen chunks are now buffered and committed as a single turn on `speech_final` (true end of speech), with `utterance_end_ms` as a backstop for when background noise stops endpointing from firing, and a flush on stop so a trailing sentence isn't dropped.
- **System audio is now actually captured on macOS.** The Chromium shipped in Electron 33 had no macOS system-audio capture at all, so `getDisplayMedia` handed back a stream with no usable audio: on headphones the interviewer was never transcribed, and on speakers their speech reached only the microphone and was labelled **You**. The app now runs on Electron 44, whose Chromium captures system audio through Apple's CoreAudio Tap API (macOS 14.2+), and declares the `NSAudioCaptureUsageDescription` permission that API requires. The microphone also runs with echo cancellation so speaker output no longer bleeds into the candidate channel, and a system stream that arrives without an audio track now raises a visible error instead of silently degrading to mic-only.
- System-audio capture without Screen Recording permission now fails quietly with a clear prompt instead of throwing `Failed to get sources` / `Video was requested…` unhandled rejections. When the permission is already denied, the app skips the capture attempt and guides you to grant it (continuing mic-only).

### Changed
- **Sentence pause is now configurable** under Settings → Sentence pause: Short (0.5s), Normal (0.9s, the new default) or Long (1.5s) of silence before a sentence is treated as finished. The previous behaviour was a fixed 0.3s, which cut off anyone who paused to think. Auto-answer's own debounce dropped from 1.3s to 0.6s to compensate, so the time from question to answer is about what it was.
- **Academic answer style**: answers are now flowing spoken prose (about 3–8 sentences) instead of the previous outline form with `- ` bullet points, and are cast as a dissertation-defense reply rather than a job-interview reply. The prompt now treats your Knowledge Base as the authoritative source, reproduces numbers, formulas, algorithm names and experimental conditions verbatim, labels anything drawn from general knowledge, reports contradictions between sources, and refuses to invent facts or references. Question extraction targets a committee member's current question, dropping preambles and preferring a genuine follow-up over an already-answered question.
- **Answer length default raised** from 500 to 1200 characters, since character count is now a hard ceiling rather than a target. Existing installs still on the old 500 default are migrated automatically (a length you chose yourself is kept); adjust it any time under Settings → Max characters.
- Renamed the app (display name) to **Real Time Interview Copilot**. The npm package, bundle id (`com.interview.copilot`), repo slug, and local data directory are unchanged, so existing settings/keys/JD/Knowledge Base are preserved.

### Added
Expand Down
2 changes: 1 addition & 1 deletion CONTRIBUTING.md
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,7 @@ npm run lint # ESLint
npm run format # Prettier (write)
```

Use the Node version in [`.nvmrc`](.nvmrc) (`nvm use`). Node ≥ 18 is required.
Use the Node version in [`.nvmrc`](.nvmrc) (`nvm use`). Node ≥ 22.12 is required — `@electron/packager` refuses to run below it, so packaging fails on older Node even though `npm install` only warns.

## Project layout

Expand Down
22 changes: 12 additions & 10 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -4,11 +4,11 @@
[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)
![Platform](https://img.shields.io/badge/platform-macOS%20%C2%B7%20Windows%20%C2%B7%20Linux-lightgrey)
![Node](https://img.shields.io/badge/node-%E2%89%A518-brightgreen)
[![Providers](https://img.shields.io/badge/LLM-DeepSeek%20%7C%20Gemini%20%7C%20OpenAI%20%7C%20Ollama-c15a37)](#features)
[![Providers](https://img.shields.io/badge/LLM-DeepSeek%20%7C%20Gemini%20%7C%20OpenAI%20%7C%20Anthropic%20%7C%20Ollama-c15a37)](#features)

**Hear the interviewer's question and draft your answer — live, and entirely on your machine.** Built for interview practice, mock interviews, and self-review.

> Real-time interview assistant — live dual-channel transcription (Deepgram) + LLM answers (DeepSeek / Gemini / OpenAI / Ollama), grounded in your own documents.
> Real-time interview assistant — live dual-channel transcription (Deepgram) + LLM answers (DeepSeek / Gemini / OpenAI / Anthropic / Ollama), grounded in your own documents.

A single-window Electron desktop app. It listens to a conversation, separates **you (microphone)** from the **interviewer (system audio)**, and on a hotkey detects the interviewer's current question and drafts a concise, first-person answer — using the last ~15 turns of dialogue plus any documents you upload as context.

Expand Down Expand Up @@ -36,7 +36,7 @@ Using real-time answer generation during a **live** interview without the other
This app sends data to third-party APIs **only for the providers you configure**:

- **Audio** (microphone + captured system audio) is streamed to **Deepgram** for transcription.
- **Transcripts, your question, and uploaded document text** are sent to your chosen **answer provider** (DeepSeek / Gemini / OpenAI) to generate answers.
- **Transcripts, your question, and uploaded document text** are sent to your chosen **answer provider** (DeepSeek / Gemini / OpenAI / Anthropic) to generate answers.
- API keys are stored **locally only**, in `app.getPath('userData')/settings.json` (e.g. `~/Library/Application Support/interview-copilot/settings.json` on macOS). They are never committed to the repo or sent anywhere except the provider's own API.

Want **zero cloud**? Use **Ollama** as the answer provider (runs locally). A fully local STT option (Whisper) is on the [roadmap](#roadmap).
Expand All @@ -50,15 +50,15 @@ Want **zero cloud**? Use **Ollama** as the answer provider (runs locally). A ful
- ⚡ **Auto-answer (optional)** — flip the toggle and it answers on its own as soon as the interviewer finishes a question, no keypress needed.
- 🧠 **Context-aware** — answers are grounded in the last ~15 turns of dialogue + your uploaded résumé / JD / notes (a persistent Knowledge Base you can add to, update, and clear).
- 🎯 **JD customization** — paste or upload a job description in Settings; it's persisted and tailors every answer to the target role.
- 🔌 **Switchable providers** — DeepSeek, Gemini, OpenAI, or local Ollama, with automatic retry + model fallback.
- ✍️ **Concise, outline-style** answers (default ≤ 500 chars), shown alongside the detected question so you can edit and regenerate.
- 🔌 **Switchable providers** — DeepSeek, Gemini, OpenAI, Anthropic, or local Ollama — pick the model from the main window.
- ✍️ **Spoken academic prose** — answers are continuous sentences (roughly 3–8, default ceiling 1200 chars), never bullet lists, grounded in your uploaded materials with figures, formulas and terminology reproduced verbatim; shown alongside the detected question so you can edit and regenerate.
- 🖥️ **Single, clean UI** — live transcript on the left; question / answer / knowledge base on the right.

## Requirements

- **macOS 13+**, **Windows**, or **Linux** — runs on all three; system-audio capture is most seamless on macOS (see [Platform support](#platform-support)).
- **Node.js ≥ 18** (Node 20 LTS recommended; see `.nvmrc`).
- API keys: [Deepgram](https://console.deepgram.com/) (STT, required) and at least one answer provider — [DeepSeek](https://platform.deepseek.com/) / [Gemini](https://aistudio.google.com/apikey) / [OpenAI](https://platform.openai.com/api-keys), or a local [Ollama](https://ollama.com/) install.
- **macOS 13+**, **Windows**, or **Linux** — runs on all three. Capturing the interviewer's audio needs **macOS 14.2+**; on macOS 13–14.1 use the [virtual-device fallback](#platform-support).
- **Node.js ≥ 22.12** (see `.nvmrc`) — required by `@electron/packager`, which builds the macOS app.
- API keys: [Deepgram](https://console.deepgram.com/) (STT, required) and at least one answer provider — [DeepSeek](https://platform.deepseek.com/) / [Gemini](https://aistudio.google.com/apikey) / [OpenAI](https://platform.openai.com/api-keys) / [Anthropic](https://console.anthropic.com/settings/keys), or a local [Ollama](https://ollama.com/) install.

## Quick start

Expand Down Expand Up @@ -107,7 +107,7 @@ After that it launches normally. (Building from source — `npm run package` —

| Platform | Run it | Mic (candidate) | System audio (interviewer) |
|---|---|---|---|
| macOS 13+ | `make run` (builds & opens the app) or a `.dmg` | ✅ | ✅ Screen-Capture loopback — grant Screen Recording |
| macOS 13+ | `make run` (builds & opens the app) or a `.dmg` | ✅ | ✅ on **macOS 14.2+** via CoreAudio Tap — grant **Screen & System Audio Recording**, and allow system-audio recording when macOS asks. macOS shows a screen-recording indicator even though only audio is captured. On macOS 13–14.1 use the virtual-device fallback below. |
| Windows | `…Setup.exe`, or `npm install && npm start` / `make run` | ✅ | ✅ `getDisplayMedia` loopback (tick "share system audio") |
| Linux | `.AppImage`, or `npm install && npm start` / `make run` | ✅ | ⚠️ no `getDisplayMedia` loopback — pick a PulseAudio/PipeWire **Monitor** source under **Interviewer audio** |

Expand All @@ -127,6 +127,8 @@ src/
prompt.js prompt builders (answer + question extraction) — pure, tested
llm.js provider-agnostic retry + fallback
gemini.js Gemini REST (SSE)
anthropic.js Anthropic Messages API (SSE)
models.js model picker lists (live + curated fallback)
openaiCompat.js OpenAI-compatible client (DeepSeek / OpenAI / Ollama)
store.js knowledge base (context stuffing)
documents.js txt/md/pdf/docx parsing + chunking
Expand Down Expand Up @@ -168,7 +170,7 @@ If Real Time Interview Copilot is useful to you, please **⭐ star the repo**

## 中文

基于 **Deepgram**(实时转写)+ **DeepSeek / Gemini / OpenAI / Ollama**(答案生成)的桌面应用:双声道识别(麦克风=面试者、系统声音=面试官),按 `Ctrl+A` 自动识别面试官当前的问题,并结合最近约 15 轮对话 + 你上传的资料,生成简洁的第一人称大纲式答案(默认 ≤500 字)。
基于 **Deepgram**(实时转写)+ **DeepSeek / Gemini / OpenAI / Anthropic / Ollama**(答案生成)的桌面应用:双声道识别(麦克风=面试者、系统声音=面试官),按 `Ctrl+A` 自动识别面试官当前的问题,并结合最近约 15 轮对话 + 你上传的资料,生成第一人称的学术口语化答案(连贯成段,不用要点列表;默认上限 1200 字符),数字、公式与术语严格照搬资料原文。

**⚠️ 免责声明**:本工具用于**面试练习、复盘、模拟面试与无障碍辅助**。在**真实面试**中未经对方知情使用实时答题,可能违反对方公司政策、协议、学术诚信规则或当地法律,且通常被视为不诚信行为。**如何使用、是否合规由你自行负责**,作者不为此背书,软件按「现状」提供、不附带任何担保。

Expand Down
Loading