PRATIK VERMA SR. SOFTWARE ENGINEER

AI systems & backend engineering · 4 years shipping

Pratik Verma

I build the AI systems that still have to work on Monday morning.

Most AI demos fall over the first time real users show up. I build the other kind. The platform I architected last year places 300+ automated medical interview calls a day over live telephony — forty-odd questions each, nine hours straight — and has the reports filed before anyone thinks to ask for them.

Download the PDF résumé Start a conversation

01

The four numbers I'd want you to check

Each one came out of a system that is still in production today. Ask me how any of them were measured and I'll walk you through the instrumentation.

  • 0+

    interviews placed daily

    Automated medical tele-interviews running across a nine-hour operational window, every working day, on live carrier telephony.

  • 0%

    OCR extraction accuracy

    Document intelligence with a human-in-the-loop gate: the model reads everything, a person only sees what the rules layer flags.

  • 0%

    less manual document handling

    What used to be someone reading pages front to back is now someone approving exceptions. Same team, different job.

  • 0%

    faster API responses

    Won back through query tuning and honest service boundaries under production load — not by renting a bigger instance.

02

Selected work

One project, told properly. It's the clearest picture of how I work: own the architecture, integrate the messy parts, and stay on it after it ships.

Insurance · Healthcare · Production since 2025

BGIL — enterprise AI medical interview platform

My role Architecture, AI orchestration, backend, deployment

Before an insurer underwrites a policy, someone has to get a medical statement out of the applicant. Traditionally that means a nurse or a call agent working through forty-odd questions on the phone, then typing the whole thing up afterwards.

BGIL makes the call instead. An AI voice agent dials out over Exotel SIP, runs the interview through LiveKit Cloud with live LLM orchestration, and hands the session off the moment it's done. n8n picks up from there and turns the transcript into what the business actually needs — a structured CSV, a formatted Medical Examination Report, the MIS row, the recording link — routed to the right person without anybody chasing it.

It carries 300+ interviews a day across a nine-hour window. I owned it end to end: voice orchestration, telephony integration, backend workflows, reporting automation, and the production environment all of it runs on.

  • Questions per session40–45
  • Daily load300+ users
  • Operating window9 hours
  • Post-call artefactsCSV · MER · MIS
Outbound dialExotel SIP
Live sessionLiveKit Cloud
Interview agentLLM orchestration
StructuringTranscript → schema
Post-calln8n automation
DeliveryMER · CSV · MIS

Enterprise RAG retrieval

Retrieval-augmented pipelines over internal knowledge so people stop opening six documents to answer one question. Built for grounding and traceability, not for a demo video.

Human-in-the-loop validation

Automated document intelligence paired with a deterministic rules layer. The model proposes, the rules flag, a person decides. That combination is what got accuracy to 99.8%.

Paid-content protection

DRM video protection plus Content Security Policy headers on an EdTech platform, closing off unauthorised redistribution and the XSS surface that came with it.

Release plumbing

Docker environments and GitHub Actions pipelines that turned deployment from a scheduled event into something you do on a Tuesday afternoon.

03

Track record

Four years, three companies, one direction of travel: payments → full-stack product → owning AI systems in production.

  1. Jul 2025Present

    Senior Software Engineer — Nu10 Technology

    • Led architecture and delivery of production Agentic AI platforms, owning the calls from system design and orchestration through to what actually runs on the servers.
    • Designed scalable RAG architectures for enterprise knowledge retrieval, so answers come with their source instead of a search box.
    • Shipped AI voice agent workflows on real telephony: live LLM orchestration, SIP, and automated post-call reporting.
    • Built human-in-the-loop validation combining document intelligence with rule-based verification for operational accuracy.
    • Set up Docker environments and GitHub Actions CI/CD, cutting manual deployment work and making releases repeatable.
    • Cut API response latency 35% under production load through service boundaries and database work.
    • Translated enterprise workflows into architectures with product teams and stakeholders who don't speak in endpoints.
  2. Jul 2023Feb 2025

    Full Stack Developer — Axon Aio Technologies

    • Built FinTech and EdTech platforms end to end: backend architecture, API design, frontend engineering, security, and cloud deployment.
    • Wrote scalable REST APIs and modular Node.js/Express services behind transaction, administrative, and customer-facing workflows.
    • Implemented DRM video protection and CSP controls, shutting down unauthorised content distribution and XSS exposure.
    • Ran production services on AWS EC2 and S3, including the deployment path that got code there.
    • Drove implementation across frontend, backend, and infrastructure with cross-functional stakeholders.
  3. Aug 2022Jul 2023

    Junior Software Developer — Trustly Pay Pvt Ltd

    • Built backend services and React operations dashboards for FinTech payment workflows, covering transaction processing and merchant administration.
    • Developed PayIn, PayOut, checkout, merchant onboarding, and transaction management for an end-to-end digital payments platform.
    • Implemented secure authentication and Role-Based Access Control across merchant and administrative surfaces.
    • Tuned queries and data-access patterns to bring down overhead on transaction-heavy operations.

04

Capabilities

Everything listed here has been through a production deployment. Nothing here is from a tutorial.

Languages
PythonJavaScriptTypeScript
Backend
FastAPINode.jsExpress.jsREST APIsScalable API design
AI engineering
LLM applicationsAgentic AILangChainLangGraphRAGMulti-agent systemsPrompt engineeringOCR pipelinesHuman-in-the-loop
AI infrastructure
MCP serversWebMCPn8n automationLiveKit CloudExotel SIP
Frontend
React.jsTailwind CSSOperational dashboards
Cloud & DevOps
AWS EC2AWS S3Google CloudDockerGitHub ActionsCI/CD
Architecture
System designMicroservicesDistributed systemsMonorepoTurborepo
Security
RBACABACCSP headersDRM protectionRate limiting
Data
Query optimisationSchema designTransaction-heavy workloads

05

Hiring for something that has to hold up in production?

I'm open to senior engineering roles in AI systems and backend platforms. Email is the fastest way to reach me — I reply within a day, and I'll tell you straight if it isn't a fit.

Emailpratikverma9691@gmail.com Phone+91 87708 00807 GitHubelliot09alderson LinkedInpratik-verma
Download the PDF résumé

B.Tech, Computer Science · CSVTU · Graduated 2021