The video opens with a simple but provocative question: what if the best trade is the exact opposite of what everyone says? The idea is to build an AI bot that flips the long calls shared as sure winners by crypto influencers into shorts. The whole experiment runs on Binance testnet across a 48-hour window, with results shared step by step.
A clear frame is drawn first: there is no such thing as zero risk on an exchange; risk starts the moment money hits the table. That is why the system runs on play money over the test network and is explicitly not investment advice. This framing turns the rest of the experiment from a gambling story into an engineering trial.
The architecture has three modules: pull data from a Telegram channel, parse and analyze the signal with Gemini, then send the order to the exchange. The original plan was to slip into Telegram groups and pull information from inside, which required renting a virtual number. But the bank blocked the Nigeria-bound payment, and the alternative service had a discouraging minimum deposit.
The fix is switching to a scraping setup that needs no account and no login, collecting only public data. A public Telegram group is found through research, and the shared signal texts are pulled from there. Once the Telegram connection is up and the latest message in the channel is captured, the data leg is complete.
For the analysis leg, an API key is created through Google AI Studio, offering up to 500 questions a day. The first attempt fails with a model-name error; after opening the docs and copying the current model name, the AI analysis starts returning answers. From then on, every signal passes through the language model before becoming a trade.
On the execution leg, an API key is generated on Binance so the bot can talk to the exchange. First a gateway error appears; a quick search reveals the testnet address has moved, and updating it helps. Then a timestamp error shows up and is resolved by syncing the computer clock. The connection succeeds, and a first trial position appears on the panel with a small loss.
Then the three modules are merged into one main script: when run, it finds five messages in the channel, produces a strategy, and forwards the order; the position is confirmed on the Binance panel. Watching it through the terminal is fun, but a simple Streamlit dashboard is built to see the data in one place, tracking signal capture, strategy output, and order flow.
The interim results raise the stakes: first about $130 in profit and 4 percent return, then roughly 12 hours later the tally climbs to $462, around the 15 percent band. When a new signal is caught, a reverse-direction position is opened and the wait begins; a long call on Bitcoin, for instance, is read by the system as selling pressure.
The final tally reaches $776 in profit, stressed in the video as the product of a single channel's signals. A natural question follows: if one channel did this, what would a system tracking and cross-checking a hundred channels do? The idea is left to the comments, opening the door to the next experiment.
The closing keeps a realistic note: zero risk exists only on the demo network, and the experiment was made to test an idea in a controlled way, not to sell anyone a strategy. The full code, along with the Gemini prompts and the architecture, is shared on GitHub, and viewers are asked to report bugs in the comments. The next experiment is teased too: a small server left open to the internet to see who notices it first.
AI commentary
"I think the real value of this experiment is not the profit figure but the method: instead of debating the fade-the-crowd idea, it was made measurable on a testnet. I would run this bot over a much longer window before letting it anywhere near real money."
AI assessment
To steelman the other side: betting against influencer signals may be more rational than it sounds. Joining the long crowd at peak euphoria means buying an overheated market at the top; as Bookmap's analysis stresses, herd psychology produces market extremes, and positioning against those extremes is the textbook definition of contrarian strategies. So the experiment's intuition is not hollow.
But the methodological limits speak louder than the results. The sample is 48 hours and a single channel; the market regime, signal count, position sizes, and starting capital stay unclear. The heavier problem is the gap between testnet and live markets: as the Paybis guide underlines, fees, slippage, and latency erode paper profits in reality; thin order books and compounding commissions on every trade can flip the outcome of a frequently trading bot.
On verifiability the picture is incomplete: what $776 means relative to the starting capital, whether the signals themselves were bad or the inversion saved them, and how losing trades were distributed all remain unknown. There is no independent replication, and one experiment cannot generalize. The figures should be read as an observation to be repeated, not as proof.
My practical takeaway: for someone learning and tinkering, this is a great project template on a testnet; not for real money. If I ran a similar bot, I would stretch the window to weeks, track several channels at once, book realistic fees and slippage on every trade, and measure the raw signal quality separately from the inversion. I would not draw conclusions before seeing that table.
Sources
6 links; no other published story cites them. Stories sharing a link do not confirm each other; a source's origin is not inferred from how often it is cited.
- @youtube Niyazi Can Çinkır — bölüm videosu
- @bookmap https://bookmap.com/blog/going-against-the-crowd-in-trading-contrarian-strategies-for-market-success
- @paybis https://paybis.com/blog/how-to-backtest-crypto-bot
- @binance https://www.binance.com/en/square/post/302385715081681
- @gainium https://gainium.io/crypto-strategy-testing
- @coinbureau https://coinbureau.com/guides/how-to-backtest-your-crypto-trading-strategy
contra-trading · binance testnet · telegram signals