Recommended Free Tools
InterviewBuddy is a mock interviewer for software-engineering interviews, built to help one nervous learner practice behavioral, technical-concept, and system-design questions. In a 2026 DEV Community post, its creator, Isha Vishwakarma, describes fine-tuning Qwen3-8B to deliver structured feedback and carry brief coaching notes between sessions. The account explains how the prototype was assembled; it does not show that the coach improves real interview outcomes.
What InterviewBuddy is designed to do
Vishwakarma built InterviewBuddy for her friend Rose, who was preparing for software-engineering interviews. “Patient” describes the intended manner of a supportive coach for a nervous learner; this is not a tool for clinical patient interviews.
As an Amazon Associate I earn from qualifying purchases.
A practice turn works like a chat: the learner responds to an interview question, and the application sends that answer to a fine-tuned model. The response is meant to provide feedback and then ask the next question. The author describes three practice areas:
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problems- Behavioral questions
- Technical concepts
- System design
The post says the coach is intended to respond to moments such as “I’m blanking” with a hint rather than simply giving away the answer, reassure a nervous learner, and help trim a rambling response. These are reported design goals and behaviors, not proof that every generated reply follows them reliably.
#1 Best Overall
- Careercup, Easy To Read
- Condition : Good
- Compact for travelling
How the coaching response is structured
The author says the model was trained to organize feedback into four parts:
- What worked — identify a useful element in the learner’s answer.
- To improve — point out what could be clearer or stronger.
- Stronger version — retell the learner’s own story in a more effective form, rather than inventing a new experience.
- Next question — continue the practice session.
That format makes the intended interaction more specific than a generic chatbot exchange: it combines critique, a model of how to express the same experience, and a continuing interview prompt. Whether the rewritten version consistently preserves the learner’s facts is not independently established in the post.
What the reported system uses
Vishwakarma describes a browser chat interface backed by a Flask application. The language model is Qwen3-8B, adapted with a rank-32 LoRA through Tinker. For continuity, Backboard is described as a separate service that stores and retrieves short notes between sessions.
At the start of a session, the application reportedly retrieves earlier notes; at the end, it writes a summary containing two strengths and one or two areas to practice next. The post reports that 27 of the 105 conversations included past-session notes, with examples designed to let the coach revisit a known weakness.
This separates two jobs: the model produces the coaching exchange, while the memory layer supplies selected context from earlier sessions. The post does not describe an audit of the full application’s data flow. The author says the adapter could be downloaded and run locally, presenting that as a privacy benefit; this is a stated possibility, not a guarantee that all application data stays local or that a deployment meets a particular privacy standard.
How the author reports preparing and training the data
The project’s reported dataset contained 105 multi-turn mock-interview conversations. Vishwakarma says the conversations were drafted with AI assistance and then reviewed and validated by her. A validation script checked formatting, alternating roles, and the exact system prompt.
| Reported dataset detail | Figure in the 2026 post |
|---|---|
| Behavioral conversations | 30 |
| Technical-concept conversations | 25 |
| System-design conversations | 20 |
| “Hard moments” conversations, including nervousness, not knowing an answer, rambling, or asking directly for the answer | 30 |
| Total conversations | 105 |
| Training and held-out split | 95 for training; 10 held out for testing |
| Graded-feedback turns | 153 |
| Training data volume | About 150,000 tokens |
All figures in the table are the author’s project-reported counts, not independently verified statistics.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
The post reports a rank-32 LoRA adapter trained for five epochs, with loss calculated only on interviewer turns so the model would learn the coach’s role rather than imitate candidate answers. It says held-out test loss was monitored, but its descriptions of the training schedule differ: one section reports five epochs, while another describes measuring held-out loss after each of three epochs. The account does not resolve that inconsistency. It also says the run used a small fraction of the author’s credits, without stating a monetary total.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the reported evaluation does—and does not—show
A held-out set of 10 conversations and monitored test loss offer a limited check on training behavior. They do not establish that InterviewBuddy gives consistently useful coaching or helps learners perform better in actual interviews. The post provides no quantitative coaching scores, baseline comparison, independent expert ratings, or measured change in real-world interview performance.
Vishwakarma summarizes the motivation this way: “With open weights I trained the behavior in: the 4-part feedback format, refusing to give away answers, and the patience when someone is nervous.” That describes what she sought to train into the model, not an independently demonstrated level of reliability.
What this build report is useful for
InterviewBuddy is a concrete example of using an open model adapter for a narrow coaching role, pairing it with a web chat application and a distinct memory layer. Its reported data also reflects a useful design choice: practice examples cover not only interview topics, but difficult learner moments that affect how a coach should respond.
For readers assessing a similar interview-practice tool, useful questions include whether it covers the relevant interview scenarios, how consistently its feedback follows a format, what controls exist for cross-session memory, where session data is processed, and whether learner outcomes have been evaluated. Those are assessment criteria, not comparisons performed in Vishwakarma’s post.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
