Recommended Free Tools
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
InterviewBuddy is a mock interviewer for software-engineering interviews—not a tool for clinical patient interviews. In a 2026 DEV Community post, its creator, Isha Vishwakarma, describes fine-tuning Qwen3-8B to give calm, structured practice feedback, with a separate memory layer for short notes between sessions. The account explains how she built it, but does not demonstrate that it improves real interview performance.
What InterviewBuddy is designed to do
Vishwakarma built InterviewBuddy for her friend Rose, who was preparing for software-engineering interviews. It covers behavioral questions, technical concepts, and system design. The goal was to make practice supportive for a nervous learner while still helping her improve—not to conduct or coach clinical patient interviews.
As described in the post, each practice turn sends the learner’s answer to a fine-tuned model, which returns feedback and a follow-up question. The browser chat interface is backed by a Flask application.
A four-part coaching response
The reported response format is:
- What worked: identify effective parts of the answer.
- To improve: point out a specific area to strengthen.
- Stronger version: help retell the learner’s own story more clearly, rather than inventing a new one.
- Next question: continue the practice session.
The author also describes design goals for difficult moments: offer a hint when someone says “I’m blanking,” respond reassuringly to nervousness, and help trim rambling answers. These are intended behaviors reported by the creator, not evidence that every response consistently follows them.
#1 Best Overall
- Careercup, Easy To Read
- Condition : Good
- Compact for travelling
How the model and memory are put together
The post says the coach uses Qwen3-8B with a rank-32 LoRA adapter trained through Tinker. LoRA is an adapter-based fine-tuning approach: rather than describing a wholly new model, the account describes adjusting a relatively small set of added parameters to steer the base model toward the interviewer role.
Memory is handled separately. Vishwakarma says Backboard stores and retrieves short notes between sessions. At the start of a session, the application reads prior 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 help the coach revisit a known weak spot.
This arrangement separates the model’s general coaching behavior from a brief record of an individual learner’s progress. The post does not establish that the notes capture every useful detail or that the system’s data flow was independently audited.
What the training set and process included
Vishwakarma reports creating 105 multi-turn conversations and using 95 for training while holding out 10 for testing. The set was divided among practice types and challenging learner moments:
| Conversation category | Count reported |
|---|---|
| Behavioral interviews | 30 |
| Technical concepts | 25 |
| System design | 20 |
| Hard moments, such as uncertainty, nervousness, rambling, or asking for the answer | 30 |
The author also reports 153 graded-feedback turns and about 150,000 training tokens. She 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. These are figures and process details from the creator’s 2026 account, not independently verified measurements.
Training the interviewer rather than the candidate
The post says the rank-32 LoRA adapter was trained on Qwen3-8B for five epochs, with loss calculated only on interviewer turns. The stated rationale was to teach the model how to act as the coach, rather than train it to imitate candidate answers. The author also reports monitoring held-out test loss, but describes the measurement schedule inconsistently: one passage says five epochs, while another says held-out loss was measured after each of three epochs. The account does not resolve that discrepancy.
What the reported evaluation does—and does not—show
A held-out set of 10 conversations and test-loss monitoring can indicate whether the model fits examples beyond its training conversations, but they do not by themselves show that its coaching is helpful in practice. The post does not provide quantitative coaching scores, a baseline comparison, independent expert ratings, or measured changes in learners’ interview performance.
That distinction matters: a model can follow a requested format or produce plausible feedback without making a learner more prepared. The article is best read as a first-person build report describing an approach and its intended behaviors, not as proof of effectiveness.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What local availability means for privacy
Vishwakarma says the adapter could be downloaded and run locally, which she presents as a privacy benefit. Local execution may offer a deployment option that keeps some processing on a user-controlled machine, but the post does not establish that the complete application runs locally, that all session data stays on-device, or that the system meets a particular privacy standard. Backboard is described as a separate memory layer, and the account does not audit the full flow of conversation data.
How to assess a similar interview coach
InterviewBuddy’s account suggests useful questions to ask about any AI practice tool. These are evaluation criteria, not a comparison of products:
- Scenario coverage: does it support the interview types you need, such as behavioral, technical-concept, or system-design practice?
- Feedback consistency: does it give specific, actionable feedback and distinguish a hint from a full answer?
- Memory controls: can you inspect, correct, or delete notes carried between sessions?
- Deployment and data handling: where are prompts, answers, and memory stored, and can you verify what stays local?
- Learner evidence: has the coach been assessed by people other than its creator, and were improvements measured against a baseline?
Source
Isha Vishwakarma’s 2026 DEV Community article about building InterviewBuddy is the source for the project-specific details and reported figures in this article.
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 problemsQuick 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.
