Tushar Mishra — Voice & Agentic AI Engineer

What I do best

Voice pipelines that work in real conversations.

Turn-taking, barge-in handling, sub-1.5s latency, eval setup, and clean state management with LangGraph. I've shipped this end-to-end as a solo founder, as a contractor for fintech voice AI, and on third-party platforms like Bolna.

Selected work

What I've shipped.

Braagi

Solo built · Live in production Flagship

A voice AI interviewer that conducts adaptive mock interviews based on the candidate's resume and the job description. Full pipeline from STT to LLM reasoning to TTS, with classification and scoring nodes that adjust questions in real time based on candidate answers.

Architecture: AbstractNode + AbstractTool base classes. LangGraph nodes for classify_answer, evaluate_and_score, decide_next_question, generate_report. Firebase JWT auth via HTTP-only cookie middleware. Credits-based pricing on Razorpay + PayPal.

LangGraph FastAPI Deepgram ElevenLabs Sarvam Gemini Redis Firebase Next.js

Call into the product at braagi.com →

Sei AI · Wise Support voice agent

Take-home, fintech voice AI

Built a Twilio-based voice agent for Wise customer support flows. LangGraph state machine handling support intent classification, knowledge retrieval, and escalation paths. Got positive feedback from CTO Ram Venkataraman (ex-Wise, ex-PayPal, ex-Amazon).

LangGraph FastAPI Redis Twilio

Bolna · Lead qualification agent

Production agent on Bolna's platform

A Naukri-style lead qualification voice agent that screens job seekers, handles intent and rejections gracefully, and routes qualified leads. Built as part of Bolna's Solutions Engineer evaluation.

Forge My Resume

Solo built · Live

An agentic resume builder with persistent memory across sessions. Remembers every input the user provides and uses it to construct ATS-optimized resumes tailored to specific job descriptions.

LangGraph FastAPI Gemini Next.js

forgemyresume.com →

Production scale

Backend systems with real users.

Digikagaz

Backend Engineer · 2 years

Indian legal-tech company. Two production products that I owned key parts of:

Legalbaaj — Legal case repository with 20M+ records, optimized for sub-millisecond read latency. Owned the read path, query optimization, and indexing strategy.

Digikagaz — Document automation tool. Interview-flow legal document generation that turns a few clicks into a complete legal document.

Box manufacturing ERP

Current client · 20% equity

Building a complete factory management system for a corrugated box manufacturer. Inventory, reel stock, production planning, quality control automation. Currently shipping in production with the client running it daily.

FastAPI Postgres Google Apps Script

India Post automation

Tool for legal professionals

A small but specific tool: takes a tracking number, monitors delivery status, captures a screenshot when delivered, and emails the proof to the lawyer. Stores case + package metadata for audit purposes.

How I work

Fixed scope. Fixed price. Source code yours.

4–6 week voice agent build

Pipeline (STT + LLM + TTS), one integration (Twilio or web), basic eval suite, deployed and callable. Source code yours.

Specialized vertical agent

Custom voice agent for one specific industry — healthcare, finance, recruiting, support. Useful when your team is shipping platform features and a customer needs something specific.

Bridge contractor

Temporary capacity while you hire full-time for senior voice/agentic AI engineering. Often a 2-month engagement that bridges the gap.

Stack

What I build with.

Voice & Agentic: LangGraph · Deepgram · ElevenLabs · Sarvam · Twilio · Vapi-compatible patterns
Backend: FastAPI · Python · Postgres · Redis · Firebase
Frontend: Next.js · React · TypeScript · Tailwind
LLMs: Gemini · Claude · GPT-4o · open-source via Ollama
Infra: DigitalOcean · Vercel · GitHub Actions