SJSuraj Jha
Let’s talk
Resume

Building with AI

Suraj Jha — AI Engineer.

I’m Suraj Jha, an AI Engineer. I build real-time voice agents, RAG systems, MCP integrations, and agentic workflows — with the backend to run them reliably at scale.

Focused onGenerative AI, Agentic AI, MCP, RAG, LLM Systems, Real-Time Voice AI

See my work Resume
Portrait of Suraj Jha
10K+daily call capacity
Voice AI
AI Engineer · Probus Insurance

  • 10K+

    Daily call capacity across concurrent bots

  • 74+

    Partner APIs monitored in real time

  • 15+

    Insurance partners integrated in the matching engine

  • 7

    Systems built, each with a full case study

About

AI that actually works in production.

I'm an AI Engineer specializing in Generative AI, LLM systems, and real-time Voice AI for insurance operations. My work spans LLM orchestration, RAG, MCP integrations, agentic workflows, and the real-time speech and backend engineering needed to run them reliably at scale.

  • Generative AI
  • Large Language Models
  • Real-Time Voice AI
  • RAG
  • Agentic AI
  • MCP (Model Context Protocol)
  • Ollama
  • Prompt Engineering
  • Function Calling
  • Structured Outputs
  • Multi-Provider AI
  • Python
  • FastAPI
  • Microservices
  • Data Processing
  • Production AI Systems

Right now

Building with AI

Building AI systems for insurance operations — a configurable voice platform, multi-provider LLM orchestration, vehicle-data automation pipelines, and MCP-based agentic workflows.

India

—:—IST

Local time — a good moment to say hi.

Daily drivers

Experience

Where I’ve been shipping.

AI Engineer

Probus Insurance Broker Private Limited

Aug 2025 — Present

Building AI systems for insurance operations — a configurable voice platform, multi-provider LLM orchestration, vehicle-data automation pipelines, and MCP-based agentic workflows.

What I did

  1. 01

    Architected a configurable AI voice platform for insurance operations, designed for 10,000+ daily calls across concurrent bots with a real-time STT → VAD → LLM → TTS pipeline.

  2. 02

    Built a provider-agnostic LLM orchestration layer spanning multiple providers (OpenAI, Anthropic Claude, Google Gemini, Groq, Mistral, DeepSeek, and others) with automatic health-based failover and live mid-call provider switching.

  3. 03

    Developed vehicle data mapping and normalization pipelines with rule-based scoring and insurance-domain validation logic for GCV and PCV quotation workflows.

  1. 04

    Designed MCP server architectures and n8n automation pipelines orchestrating LLMs, REST APIs, and PostgreSQL/Redis databases into scalable production workflows.

  2. 05

    Built RESTful microservices with FastAPI and Flask; automated data extraction from 74+ insurance partner portals using Selenium and async HTTP with rate-limit-safe execution.

  3. 06

    Operated production AI systems serving thousands of daily user interactions with high availability, using async circuit breakers, Redis-based distributed state, and Prometheus + Grafana observability.

Selected work

Things I’ve built.

Seven systems — from production voice AI to data engines. Every one comes with a full case study: the problem, the architecture, the trade-offs.

01AI / ML

AI Voice Calling Platform

A configurable voice AI platform for insurance operations, designed for 10,000+ daily calls across concurrent bots.

  • 10K+daily call capacity
  • <1sturn latency
  • ~80%of turns answered in under 50ms
  • 6LLM providers with auto-failover
  • Voice AI
  • Real-Time Systems
  • Telephony
  • LLM Orchestration
Read the case study

02Data

Vehicle Catalogue Matching Engine

A fuzzy-matching pipeline that standardizes vehicle master data across 15+ insurance partners, powering the company's quotation portal.

  • 15+insurance partners on one codebase
  • Days → minsper-partner reconciliation
  • Data Engineering
  • Fuzzy Matching
  • Data Reconciliation
  • Automation
Read the case study

03Backend & Full-stack

Conversational Quote Assistant

A chat-based front end that replaces a multi-field insurance quote form with a guided conversation.

  • 0backend changes needed
  • Chatreplaces a multi-field form
  • Conversational UX
  • State Machines
  • API Integration
  • Session Management
Read the case study

04AI / ML

WhatsApp Document Verification Bot

A privacy-first WhatsApp bot that extracts and validates fields from photographed ID documents, with encrypted storage.

  • 0document content sent to external LLMs
  • Encryptedfiles and sensitive fields at rest
  • OCR
  • Document Processing
  • Privacy Engineering
  • Vector Search
Read the case study

05Backend & Full-stack

Agent Gamification Platform

A gamification and leaderboard platform for sales agents, with real-time scoring and team standings.

  • Liveleaderboard over SignalR
  • 3roles: agent, admin, master
  • Real-Time Systems
  • Gamification
  • Role-Based Access Control
Read the case study

06AI / ML

Stock Intelligence Engine

A personal project: a local app that turns any Indian company name into a full stock report, built entirely on free data sources.

  • 11news feeds merged and deduped
  • $0data cost — free sources only
  • Personal Project
  • Data Pipelines
  • NLP
  • LLM Integration
Read the case study

07Backend & Full-stack

API Health Monitor

An async platform concurrently monitoring 74+ insurance partner APIs, with status classification and multi-channel alerting.

  • 74+partner APIs checked concurrently
  • 3alert channels: email, Slack, Teams
  • Observability
  • Async Programming
  • Reliability Engineering
Read the case study

Toolkit

The stack behind the work.

The tools and disciplines I work across — from model behavior to the infrastructure around it. Tap any tool to see where I’ve actually shipped it.

AI / ML

Building intelligent systems for real-world impact

Voice AI

Real-time speech pipelines that feel like a conversation

LLMs & Agentic

Multi-provider orchestration and tool-calling workflows

Backend & APIs

Production services that run reliably at scale

Data

Storage, processing, and fast reliable retrieval

Infra & DevOps

Shipping, monitoring, and scaling what gets built

Most shipped with

Ranked by how many of the case studies here actually use each tool.

  1. Python6 projects
  2. PostgreSQL6 projects
  3. FastAPI4 projects
  4. Docker4 projects
  5. C#2 projects
  6. .NET2 projects
  7. sentence-transformers2 projects
  8. FAISS2 projects

Systems

How a voice agent thinks in real time.

Not a flowchart — the real thing, moving. Watch a call travel through the pipeline behind my voice platform, one stage at a time.

  • <1sturn latency
  • ~80%turns on the <50ms fast path
  • 6LLM providers, auto-failover
  • 500+names in the phonetics module

  1. Twilio streams the call audio in. A short greeting is pre-synthesized and played the instant the call connects, so the caller never hears dead air while the rest of the pipeline warms up behind it.

  2. Silero VAD detects when the caller is actually speaking and DeepFilterNet strips background noise from the phone line, so everything downstream works from clean speech.

  3. Streaming speech-to-text converts the caller’s audio into text in real time and hands it to the conversation layer.

  4. A deterministic state machine resolves roughly 80% of turns in under 50 ms — no LLM involved. Only genuinely off-script utterances escalate to the LLM, backed by a per-bot RAG pipeline over PostgreSQL.

  5. The reply is synthesized and streamed back to the caller. A custom phonetics module covering 500+ names keeps Indian names pronounced correctly across regional languages.

  6. The multi-tenant control-plane dashboard (Angular + .NET) gets live updates pushed to the browser in real time. It only ever talks to the voice engine — never to telephony or AI providers directly.

Live callIllustrative

Illustrative conversation, written to show the stages in motion — not a recording of a real call.

Every external provider sits behind an async circuit breaker, so a struggling STT, TTS or LLM provider degrades gracefully — failing over automatically — instead of taking calls down.

Methodology

How I think about AI engineering.

  1. Understand the Problem

    Before touching a model or a framework, I define what success actually looks like — the constraints, the users, and the failure modes that matter most.

  2. Design the System

    AI features live inside larger systems. I design the data flow, the orchestration layer, and the interfaces between the model and everything around it.

  3. Build & Integrate

    I implement with production in mind from the start — typed APIs, sensible error handling, and integration points that are easy to reason about.

  4. Evaluate & Improve

    AI systems drift and degrade. I build evaluation into the loop so quality is measured, not assumed, and improvements are grounded in evidence.

Building AI systems requires more than models — it requires reliable software.

  • AI / LLM
  • Angular
  • ASP.NET Core
  • C#
  • APIs
  • Data
  • Production AI

Resume

Prefer the short version?

Two pages: experience, the systems I’ve shipped, and the stack — ready to forward.

Download PDF View online

Contact

Let’s build something intelligent.

Have a voice-AI, LLM or automation problem — or just want to say hi? Drop a message and I’ll take a look.

0/4000