ADAPTIVE TECHNICAL INTERVIEW TRAINING
Gadfly is a voice tutor for technical interview practice. You work through algorithms, system design and behavioral rounds out loud; it names what broke, remembers the weakness and brings it back until it clears.
No card · Desktop only · Chrome or Edge · Voice runs in your browser tab · Code in Java or Python · Built for preparation, never live-interview assistance
THE WHOLE LOOP IN 74 SECONDS
That was the loop. Now run it on your own interview.
Start 2 sessions freeONE REAL SESSION
Gadfly asks the next question instead of completing the reasoning for you. The goal is not to recognize a good answer after reading it. The goal is to produce and defend it yourself.
Keep scrolling — the session plays as you go ↓
That last chip is a real button. Press it.
The answer does not come. The question you were dodging does.
WHY MORE PROBLEMS ARE NOT ALWAYS THE ANSWER
During an interview you must clarify the problem, explain the plan, defend why it works, write the code and adapt when the interviewer changes a constraint.
def max_subarray(nums):
best = cur = nums[0]
for x in nums[1:]:
# restart or extend
cur = max(x, cur + x)
best = max(best, cur)
return best
Solo practice usually ends with
An interview can still break on
Gadfly trains the distance between “I know this”
and “I can demonstrate it under interview conditions.”
PRACTICE THAT CONTINUES AFTER THE SESSION
Work through a voice tutoring session or a mock round. Explain, code, draw and respond without being carried to the answer.
Close the session with evidence from what you said and did: where the reasoning held, where it broke and which behavior mattered most.
Turn the most important slip into a named weakness connected to a pattern and skill facet.
Bring it back after the answer is no longer fresh. A weakness clears after repeated clean performance, not one good evening.
Perform → Review → Diagnose → Repeat
WHY NOT JUST A CHAT
ChatGPT, Claude and Gemini are good at this: a concept you have never met, the full solution to a problem you gave up on, a warm-up on the morning of the interview, a voice mock with a small prompt. They can be asked to hold back, they can remember you, and they are free or already paid for.
The tutor here runs on the same kind of model you already talk to. What is different sits outside it: a journal the model does not write, a score that never reads your rating, and a next problem chosen from that journal.
| With a chat, you do it yourself | In a Gadfly session |
|---|---|
| Prompt it to play the interviewer and not to help. Whether it keeps to that is its call, every turn. | In an algorithms mock the hints are off by policy, the interviewer is told nothing about your past sessions, and the score is filed at the end whether or not you ask. Mock mode is in the free sessions. |
| Record it, transcribe it, and find the minute where it went wrong. | The session closes with one slip: the minute, and your words as the record has them. If the record is too thin to show one, it says nothing rather than guess. |
| Tell it how you think it went, then ask for a grade. | You rate yourself first. The coach's number is computed from the session record and never reads your rating; the model's reading counts only where it cites a moment that happened. |
| Keep a spreadsheet of which mistake came back, and decide each time whether it is the same one. | You file it under one class from a fixed list. The journal counts it, closes it after three sessions in which a problem that trains it ran green, and reopens it at zero if you file it again. |
| Schedule the repeat yourself, and pick tomorrow's problem. | A problem you rated “Did not solve” comes back on day 1, 3, 7 and 14. The next new problem is scored against your journal, and the plan says why it was picked. |
Scores, the counter and test runs are on the algorithms track. A system-design session also closes with one slip found the same way; a behavioral session closes with a scorecard of its own: four axes by fixed rules over the record, your own read first. What a session found reaches the next tutoring session on the same track for two weeks. The morning queue is the capture in the act above.
Use a chat when you need an answer tonight.
Use Gadfly when you want the next tutor session to open with what this one found.
If you already record your mocks, transcribe them, file each slip under a name, count how often it returns and pick tomorrow's problem from that count, you do not need Gadfly. If you do not, two free sessions show whether the same slip comes back twice.
Two sessions free, on a laptop in Chrome or Edge; speech models download once. Code in Java or Python. Mock mode included.
ONE GOAL · THREE ROUNDS
ALGORITHMS
Make the reasoning visible before the code appears.
Clarify the problem, explain the approach, state why it works, implement it and handle variations. Gadfly pushes on the missing step instead of accepting a correct final answer in silence.
SYSTEM DESIGN
Defend decisions, not a collection of boxes.
Start with requirements and scale. Build the system aloud, draw the architecture and defend APIs, data models, bottlenecks and trade-offs when the interviewer probes them.
BEHAVIORAL
Make your contribution clear without manufacturing a perfect story.
Practice real examples under follow-up pressure. Separate your decisions from the team’s, explain the result and say what was not measured instead of inventing a number.
All three tracks feed one journal, one skill history and one practice plan.
WHEN THE INTERVIEW DATE IS REAL
Tutor sessions ask the next useful question without handing over the solution. Mock sessions remove the hints, keep the pressure on and wait until the end to debrief what happened.
With a session pack, code review is its own room: bring a class of your own, connect a GitHub repository, or take a project from the catalog, and talk it through. The debrief judges the code together with the transcript that produced it and the edits you made while you talked — not as an isolated snippet.
Review can connect:
NOT ONLY“This branch is difficult to read.”
BUT“You said the input would remain unchanged, but the implementation sorts it in place. An interviewer will ask you to reconcile that trade-off.”
The goal is not prettier code. It is code you can explain and defend.
THE MEMORY IS THE PRODUCT
Gadfly builds a structured history of the patterns and interview behaviors measured across your sessions. A repeated slip becomes an error class to train, not another paragraph of feedback to forget.
See where the weakness lives
The skill map separates recognizing a pattern from explaining, implementing, adapting and finishing it on time.
Test it after the answer fades
A problem you struggled with comes back on day 1, 3, 7 and 14. One you found easy comes back once. You set that by rating the session, so the queue tests the skill, not your short-term memory of the solution.
Require repeated evidence
Three clean runs clear the weakness. One good attempt is progress; repeated performance is the change the plan can trust.
The next problem is not random.
It has a reason to be there.
WHY GADFLY?
Socrates was called the gadfly of Athens: persistent, uncomfortable and impossible to ignore. He did not make people think by giving them better answers. He kept asking until they could defend their own.
That is the standard for every Gadfly session: the answer should still be yours.
/ˈɡad·flʌɪ/ n. a persistent, annoying questioner — one that provokes others into thought.
TRY THE LOOP FIRST
Complete a real session and leave with one concrete diagnosis.
Stay on the track and, within two weeks, a tutoring session opens with what session one found. Change the problem and see whether the same behavior appears again.
Train or test the weakness and decide whether the resulting plan is useful enough to continue.
No card · A session counts the moment it starts · Nothing is charged today · Mock mode included · Code in Java or Python
Voice input and output run inside your browser tab, on speech models downloaded once. The privacy page explains which transcripts and session records are stored.
Gadfly is built to train you before an interview. It cannot listen to or help answer questions during a real one.
Export your data or delete the account when you choose. When a pack ends, history and export remain available.
A pack is a one-off purchase of a set number of sessions and the days to use them. There is no subscription to cancel, and each account has a daily usage ceiling you can see in Settings.
Gadfly shows what happened in your sessions. It does not pretend to know whether a company will hire you, or reduce readiness to an unsupported percentage.
Gadfly is an independent preparation product and is not affiliated with or endorsed by Google, Meta, Amazon, Microsoft or other employers.
ONE PAYMENT · NO SUBSCRIPTION
Two sessions are free, no card. After that you buy a pack of sessions and a window to use them — one payment, nothing renews, and VAT is already inside the price.
A pack ends when its sessions run out or its days do, whichever comes first.
What a pack buys is sessions. Mock mode is in the free sessions and in every pack; the packs differ in size, not in what they contain. A free mock elsewhere, or a chat asked to play the interviewer, is the right tool for one rehearsal before an onsite. What it does not leave behind is a journal that picks the next problem.
£0 · 2 sessions · no card
See the product work with your own interview performance.
£19.99 · 8 sessions · 30 days · One-off payment
Turn feedback into focused practice for your next interview.
£35.99 · 16 sessions · 60 days · One-off payment
The same product with twice the room — for a longer run-up or several rounds.
Prices include VAT — what you see is what you pay · 14-day refund, no reason needed · No recurring charge · History and export stay available when a pack ends
For scale: an hour with a live engineer starts at $179 on interviewing.io’s own checkout and reaches $399 with a named Google interviewer on Prepfully. Every one of those hours ends when the hour does, and the next one starts from zero.
No. Gadfly works before an interview: it runs practice sessions, keeps your history and refuses to hand you answers. It lives in a browser tab; the voice runs inside that same tab and touches only your microphone and speaker — there is no overlay and it does not read your screen. Using it during a real interview or assessment is against the Terms, and the thing that gives AI cheating away, not being able to speak to your own code in detail, is exactly what Gadfly trains. Some employers allow an assistant in the room; Gadfly trains for the room without one.
In a tutoring session Gadfly never confirms that an answer is right; it asks you to check it against a case. In an algorithms mock you rate yourself first, then six axes are scored by fixed rules over the record of the session; the model's reading is averaged in only where it points at a moment that actually happened, and anything it cannot anchor is thrown out. Scores are practice signals, not a verdict on you, and Gadfly never shows a readiness percentage or predicts whether you will be hired.
Gadfly is built and run by one person, Dmytro Hryshchenko — an individual, not a company — and [email protected] reaches him directly. Your account and practice history live on a server in Germany, and you can download all of it as one JSON file from Settings → Your data, or delete the account there. When a pack ends, only starting new sessions stops: your history and the export stay open.
Sessions. The loop between them is not behind the wall: the diagnosis, the journal, the drills and the next problem are in the free sessions and in every pack, and so is mock mode. A pack buys more sessions inside a window. What a free mock elsewhere does not leave behind is a journal that counts the mistake you logged and picks the next problem from it. For one rehearsal before an onsite, a free mock is the right tool. For a loop of several rounds inside a month, the eight-session pack is sized for it.
You can. For a concept you have never met, the full solution to a problem you gave up on, a warm-up before the interview or a voice mock with a small prompt, they are the right tool; they can be asked to hold back, they can remember you, and they are free or already paid for. Gadfly's tutor runs on the same kind of model; what is different sits outside it, and it files the part you would otherwise do by hand. An algorithms or system-design session closes with one slip, the minute and your words from the record, whether or not you ask, or with nothing when the record is too thin to show one. In an algorithms session you rate yourself before the coach's score appears, and that score is computed from the session record, not from your rating. The mistake you log goes under one class from a fixed list, is counted, closes after three sessions in which a problem that trains it ran green, and the next problem is scored against that journal. Use a chat when you need an answer tonight. Use Gadfly when you want the next tutor session to open with what this one found.
No. LeetCode and other question banks build pattern range, and Gadfly practises on that catalog rather than replacing it. What Gadfly trains is the demonstration: clarifying the problem, reasoning aloud, defending why a solution is correct, writing code under interview conditions and adapting when the interviewer changes a constraint. Where a catalog problem has tests, they were written for it here and checked against an independent solution before they count.
You set a goal as a company plus a level, and the practice plan follows it. Gadfly carries company programs for Amazon, Apple, Google, Meta, Microsoft, Netflix and Uber; for any other employer you still pick a level and work from the full catalog, without company-specific weighting. Gadfly is independent and not affiliated with any employer.
No. Gadfly scores what you reason and decide, taken from the text of the session, not how you sound. Speech recognition runs on your own machine and what it heard appears as text in front of you, so a misheard word is one you can retype. If speaking gets in the way, the whole session can be done by typing, with the same diagnosis and the same plan.
No — that is the case the smaller pack is for. Start with the two free sessions: they include mock mode and each one ends with a diagnosis, so within days you know what to fix. The shortest full cycle of the Gadfly loop is 14 days, because a problem comes back on day 1, 3, 7 and 14 — one pass at one weakness fits inside a fortnight. Eight sessions over 30 days is roughly three evenings a week; if you have longer, sixteen sessions over 60 days is the same product with more room.
If you can, yes, and Gadfly does not claim to replace one. A human notices things a model does not. What a paid hour does not leave behind is the loop: Gadfly keeps the journal, names the recurring weakness and schedules its return, so you walk into the next human mock knowing what to watch for.
A laptop or desktop on Linux, Windows or macOS, a current Chrome or Edge, and an internet connection — the voice is local but the tutor's thinking is not. Nothing to install: the voice runs inside the browser tab on about 1 GB of speech models that download once into the browser's own storage, turn your speech into text and read the tutor aloud on your own machine, and never upload audio. Chrome and Edge are what we test on; Firefox and Safari are not verified. Without the download the tutor stays silent and the session works in text.
Every Gadfly session can be done by typing: the tutor replies in text and the diagnosis and plan are the same. Voice is the default because interviews are spoken. Phones and tablets are not supported for sessions — the code editor needs a keyboard and a wide screen, and the in-browser voice is not built for iPad, iPhone or Android tablets.
A session is one round on one track in any mode — tutor, solve or mock — and it is the unit your free sessions and your pack are counted in. Gadfly counts it the moment it starts, so a session you abandon still counts. There is no timer inside a session: you end it when you are done. A tutoring session closes with a diagnosis and tomorrow's queue; a mock ends with the interviewer's debrief.
In tutor mode Gadfly asks the next question that moves your reasoning forward and never reveals the solution; when you are stuck you press "I'm stuck" and get one more step, up to pseudocode, but not the finished code. In mock mode it behaves like an interviewer: no coaching during the round, and the debrief waits until the end. Neither mode has a countdown — you end the session yourself.
Not as pasted problem text — Gadfly has no field for that. The catalog is 711 LeetCode problems tagged by pattern: pick one by number, title or pattern and the session opens the real statement. Your own code is a different matter — the code review room, with a session pack, takes a class of your own or a GitHub repository you connect.
Gadfly reviews the code together with the transcript that produced it and the edits you made while talking, not as an isolated snippet. It reads correctness, complexity, clarity, edge cases and whether the code matches the reasoning you gave, and it names the follow-up questions your choices invite. The goal is not prettier code; it is code you can defend.
Yes, and you can see it in Settings before you reach it. Each account has a daily usage ceiling and the service has one of its own; when either is reached, live sessions pause for the day and Gadfly says so instead of switching to canned lines. After about two hours or six sessions Gadfly also suggests stopping, but that locks nothing. Your history, plan and export stay available, and the ceiling resets the next day.
No, it is not a subscription. Two sessions are free and need no card. A pack is one payment for a set number of sessions and the days to use them — 8 sessions over 30 days for £19.99, or 16 sessions over 60 days for £35.99, VAT included — with nothing renewing and no card stored by us. A pack ends when its sessions run out or its days do, whichever comes first; nothing is charged at that point and nothing is locked. Your history, plan and export stay available, and if you buy another pack the sessions add up, each pack keeping its own window. The Terms spell this out.
Try the two free sessions first — that is what they are for. After that, writing to [email protected] within 14 days of paying gets you a full refund with no reason needed, wherever you live; only above a handful of used sessions may it be reduced in proportion. If the product never worked for you at all, that is a full refund regardless of the 14 days, and days lost to an outage on our side are added back. Gadfly sells practice, not a result: no preparation product can promise a hiring decision.
Your audio never leaves your machine — recognition and the speaking voice run inside your browser tab. The text of what you said or typed and the code you wrote go to Anthropic in the United States to produce the tutor's replies, under standard data-transfer clauses, and are never used to train models. Your account and practice history live on a server in Germany. The privacy page lists every recipient, and you can export everything or delete the account from Settings → Your data.
With an email address and a password, or with Google or GitHub; you must be 16 or older, and the free sessions ask for no card. Confirm the emailed link before your first sign-in. Gadfly writes to you only about your account: confirmation, password reset, and 30 days' notice before material changes to the Terms. To leave, open Settings → Your data, download your data if you want it, and type DELETE — that removes the account, its sign-in and any purchase record, and cannot be undone.
Pick a problem, turn on the microphone — or just type — and complete the loop from reasoning to diagnosis. The first two sessions are free.
No card · All three tracks · Desktop, in Chrome or Edge