ReachAll AI

Ashish had a market thesis. I needed to build everything.

I joined ReachAll because the voice AI market was broken. Every competitor could demo well, but when you tried thousands of calls at the same time, responses got slow, calls dropped, and costs exploded. Ashish needed a production system Fortune 500 companies could depend on. No tech, no team, no codebase existed.

I hired 4 engineers from scratch and built an engineering organization, not just a codebase

I didn't start with code. I started with people. I defined roles, ran technical interviews, and hired a team of 4 engineers. I set up the engineering foundation: code reviews, deployment pipelines, monitoring, on-call rotations. I built an engineering organization, not just a codebase.

I built the entire voice AI pipeline with responses so fast users can't tell it's AI (under 300ms, when most competitors take 700-800ms)

Most voice AI products stitch together separate services: one for turning speech into text, one for generating a response, one for turning that response back into speech. Each handoff adds delay. The industry standard was around 1,200ms a year ago, now improved to 700-800ms. Still noticeable.

I built the pipeline as a single optimized flow. Response time: under 300ms, so fast users can't tell it's AI. Human conversations have 200-500ms pauses between speakers. Our agents respond within that window. A fundamentally different product experience.

We shipped agents that speak multiple languages, AI-to-AI call transfers between specialized agents, escalation to a human when needed, intelligent phone menu navigation, thousands of simultaneous calls, and automatic post-call summaries with insights.

We found our market: screening 10,000+ blue-collar hires a month with AI that speaks their language

The platform worked, but the market was crowded. I helped find the wedge: voice AI for blue-collar hiring at scale. Large Indian enterprises like Reliance hire tens of thousands of workers monthly. A human recruiter does 6 screens a day. 65% of calls go to voicemail.

We built voice agents that screen candidates in local languages, 24/7. Time-to-first-screen dropped from 48-72 hours to under 15 minutes. We landed Fortune 500 clients and reduced onboarding costs by 3x.

Fortune 500 companies didn't just evaluate our product. They evaluated our security. We passed.

Selling to Fortune 500 is maybe 40% tech and 60% compliance. Security certifications, healthcare-grade compliance, European data protection standards, data handling, encryption, audit logs, security questionnaires that take weeks.

I baked compliance into the architecture from day one: fine-grained access controls, single sign-on so enterprises can use their existing login systems, end-to-end encryption, audit logging, data retention policies. When procurement asked "how do you handle X," we had real answers. Startups with better demos lost deals because they couldn't pass security review.

I built a caching engine that cut our infrastructure costs by 3x and made responses even faster

At enterprise scale, per-minute costs from AI providers were killing margins. I built a caching engine: standard flows, common responses, and FAQs get cached and served instantly without hitting the AI. Infrastructure costs dropped 3x. Speed improved because cached responses are near-instant.

I also optimized connection management, audio buffering, routing simpler questions to lighter and cheaper AI models, and aggressive timeout handling. Result: 99.99% uptime at scale.

Some clients couldn't send data to the cloud. I built a version that runs entirely inside their own servers.

Some clients couldn't send audio to cloud services. Healthcare companies with strict privacy requirements, financial services with data residency rules, large enterprises with internal compliance policies. I built a self-hosted deployment model using open-source components: real-time audio infrastructure, self-hosted speech recognition and voice synthesis models, packaged for clients to run in their own cloud. Audio never leaves their environment.

This became a sales weapon. "Your data never leaves your servers" beats "trust us" every time.

Results

Tech Stack