Pushkar Pandey

@GeekProxy thanks

@Pushkartwt Breaking everything twice in the first week is just the standard right of passage for building AI agents. Getting the reflection loop running is a massive milestone though, keep pushing through the chaos.

Week 1 of building my Pre-Sales AI Agent. What I shipped: → Website scraping with Playwright → Lead qualification logic → Dynamic Context Injection → Reflection loop for better response What broke: → Everything. Twice. Still going. Week 2 starts tomorrow.

My AI agent was generating generic sales scripts. Fix: added a reflection loop. Flow: generate → critique → improve Now it checks: • personalization from real data • specificity of pain points • alignment with strategy If score < threshold → rewrites only weak parts.

just cut token usage in half. each agent was getting the full data dump. it only needed a fraction of it. fixed with dynamic context injection. result: → 50% fewer tokens per call → sharper outputs → less hallucination → faster responses less data. better answers.

@dev_nam_kr good shout — showing the source next to each rebuttal is going in next sprint. no more black box.

@Pushkartwt AGENT. The pre-call intelligence wedge is solid. I would also show which source triggered each predicted objection and rebuttal so coaches can trust it fast instead of treating the agent like a black box.

Building a Pre-Sales AI Agent for coaches. It scrapes your lead's website, news. Builds their psychology profile. Predicts objections with word-for-word rebuttals. Still building. Looking for 3 coaches to test free. Comment "AGENT" if you want in. (Must be following)

Unpopular opinion: Your closing script is not the problem. Your pre-call intelligence is. You can't close someone you don't understand. You can't handle objections you didn't see coming. Know the lead before the call. The close takes care of itself. #aiagent #buildinpublic

Day 3. Built the website scraping layer today. You paste a client's URL. It pulls: → Their services and offers → Real client testimonials → Pricing signals → Their exact CTA copy #Aiagent #buildinpublic

@benjiwagn right now it's standalone — but the memory layer already stores past interactions per client in a JSON store. so the agent remembers known objections, what was said last time, and compounds intelligence across every interaction.

the objection prediction is a clever angle. most people stop at research + script. curious where the generated profiles and scripts land after the call though. do they feed back into the coach's CRM or is it standalone? feels like the real value unlock is when the agent's research compounds across multiple interactions with the same lead.

I'm building a Pre-Sales AI Agent for business coaches. You type in a client's name and website. It researches them across 6 sources. Builds their psychology profile. Predicts their objections. Writes your call script. Building in public from day 1. Follow if you want to watch.

@khalidxv1 Sure why not lets connect

Coaches don't lose deals on the call. They lose them before it. Lead books. Goes cold. Coach shows up knowing nothing. Call starts from zero. That gap has a name. Pre-sales. Nobody is fixing it properly. #Aiagent #buildinpublic

@HeyMohit_ It’s not trying to “figure someone out” from one source. It combines multiple signals (website, hiring/news, past data) and builds a probability-based profile — more like informed direction than absolute truth.

A coach gets 50 leads a month. Only 10 are qualified. They find out which 10... on the sales call. That's 40 wasted calls. 40 rounds of prep and hope. All gone because nobody qualified at opt-in. Not in 60 hours. 60 seconds. That's what I'm building.

I'm building a Pre-Sales AI Agent for business coaches. It qualifies leads the moment they opt in. No more bad-fit calls. No more chasing cold leads. No more no-shows. Building in public from day 1. Follow if you want to watch. #aiagent

#DAY6 Today I work on UI part of my PDf agent application also added a feature a user can upload only on pdf in the application #AIAgent #builtinpublic #AIイラスト︎ #CONNECT

#DAY6 Today I added the Langchain loader which is more advance then the normal text extractor and in future thinking of using more advance loader that can also extract text from OCR fomat.

#Day5 – Progress Update: • Worked on the authentication flow • Added global auth state so user login status is available across the entire app • Makes protected routes + personalized chat much easier to manage • Next: connect auth with user-specific documents #buildinpublic

@helloitsolly @SenjaHQ I will love to contribute here

I'm hiring a full-stack engineer at @SenjaHQ Contractor role $100/ hour, remote, work directly with the founder (me), get sh*t done and go No Slack, no email Here's the stack, some context Application form in my reply 💜

@akramcodez Congrazulation bro

Finally joined a UK startup as a remote SDE 🥳