I'm an AI and backend engineer focused on building systems that hold up in production. My current work centers on real-time voice infrastructure: the backend behind an AI calling platform that handles roughly 20,000 automated calls a day. Alongside that, I build RAG pipelines, LLM tooling, and applied machine learning, and I'm comfortable owning a project from the data layer and backend through to the frontend.

I care about the parts of engineering that decide whether something actually ships: latency, concurrency, reliability, and clean architecture. I enjoy turning research-stage ideas into services that run, debugging at the protocol and library level when needed, and working across the stack with a team.

• Currently building production voice AI: a self-hosted SIP softswitch, a custom media bot, and a streaming speech-AI agent
• Strong on real-time backends, RAG, and LLM systems with FastAPI, PostgreSQL, and Redis
• Comfortable across the stack, from SIP and audio framing up to a Next.js dashboard
• Open to full-time roles and freelance projects in AI/ML and backend engineering

Languages

Real-Time Voice and Telephony

AI, ML, and LLMs

Backend and Data

Cloud, DevOps, and Frontend

Backend and real-time voice infrastructure for Callixo, an outbound AI calling platform that places around 20,000 automated calls a day. AI agents hold natural conversations over the live phone network, detect voicemail versus human, book callbacks, and write structured summaries back into the CRM.

• Built a voice stack from scratch: FreeSWITCH as a self-hosted SIP softswitch, bridged through a custom PJSIP/PJSUA2 media bot into a streaming speech-AI agent (STT, LLM, and TTS over one socket), handling mu-law and PCM16 on a non-blocking 20 ms media loop
• Designed a Redis-backed account allocator with health-based auto-quarantine and a dynamic concurrency ceiling, keeping a single 4-vCPU box stable at over 100 concurrent live calls
• Diagnosed and fixed outages at the library level, including a PJSIP compile-time call cap that caused SIP 486 storms and a silently failing vendor account that was wasting around 38 percent of calls
• Scale: roughly 20,000 calls a day, about a 69 percent answer rate, multi-region dialing, and multilingual conversations

FastAPI FreeSWITCH PJSIP Deepgram PostgreSQL Redis AWS Docker Next.js

An AI system that turns browsing data into a queryable knowledge graph, extracting relationships and building interactive visualizations for semantic exploration. Winner at Hackprix Hackathon S2.

• Full-stack application with a FastAPI backend, a React frontend, and a Chrome extension for one-click data export
• A RAG-powered assistant for natural-language questions over your own knowledge base
• Interactive graph visualizations with Cytoscape.js and multiple layout algorithms

FastAPI React LangChain OpenAI Supabase pgvector Cytoscape.js

An end-to-end machine learning system that predicts equipment failures from industrial sensor data, built with a full MLOps workflow.

• Automated CI/CD with GitHub Actions, Docker, and AWS ECS/ECR
• MLflow for experiment tracking, model versioning, and automated retraining
• Reached 91.2 percent accuracy, with deployment time reduced from hours to minutes

MLflow FastAPI Docker AWS ECS/ECR Scikit-learn GitHub Actions

An LLM-driven interview platform with resume parsing, live proctoring, and voice-to-voice assessments.

• Cheat detection using YOLOv11, OpenCV, and voice analytics
• STT, TTS, and voice cloning with around 90 percent transcription accuracy
• Multi-agent assistants with persistent memory using LangGraph, OpenAI, and CrewAI
• Reduced hiring cycles by 30 percent while supporting over 1,000 daily assessments

LangChain OpenAI YOLOv11 Docker AWS EC2 Nginx

• Shipped a production voice platform handling around 20,000 AI calls a day on modest infrastructure
• Seven-plus hackathons with several podium finishes in GenAI and MLOps
• Six-plus technical articles published on AI and ML
• Co-authored a research paper on Graph-Augmented RAG systems
• 1,300-plus GitHub commits across more than 50 AI/ML and backend projects
• 250-plus LeetCode problems solved

• Lower-latency real-time media pipelines and stronger observability, including SLOs and tracing
• Multi-hop retrieval and graph-augmented reasoning
• Distributed systems patterns, including Kubernetes and load and chaos testing
• Protocol fundamentals: SIP, SDP, RTP, RTCP, NAT traversal, and codecs

I'm open to full-time roles and freelance projects in AI/ML and backend engineering, particularly work involving real-time voice, RAG, or LLM systems in production.

Email · LinkedIn · Portfolio