Open to AI engineering rolesDhaka, Bangladesh · Remote-friendly

Hi, I’m Mahir · AI Engineer

I build systems
that hold up in production.

Voice agents, multi-agent pipelines, and memory that remembers. I work end-to-end: agent graphs, RAG, GPU serving, and the UI on top.

View Work→Contact✦
./whoami
role
Product Engineer @ Obby
focus
Voice AI · Agents · RAG
degree
B.Sc. CSE · NSU
rating
CF Expert (1650)
3.86
CGPA · CSE
1650
Codeforces max
2x
ICPC Regionals
3+
yrs shipping
✦LangGraph✦FastAPI✦RAG✦ChromaDB✦PyTorch✦WebRTC✦Next.js✦CUDA✦Triton✦Voice AI✦MCP✦Knowledge Graphs✦LangGraph✦FastAPI✦RAG✦ChromaDB✦PyTorch✦WebRTC✦Next.js✦CUDA✦Triton✦Voice AI✦MCP✦Knowledge Graphs
01Featured Work

Things I've built & shipped

Systems where the hard part was the orchestration, memory, and latency, not the prompt.

Voice AI · Architecture01

Multi-Tenant Voice Agents

Production AI voice agents over SIP

Multi-Tenant Voice Agents preview

Real-time voice agents for multi-tenant SaaS at Obby. Per-tenant call routing over SIP, tenant config beside a fixed platform contract, and idempotent delivery out to CRM, email, and SMS.

  • Per-tenant resolution, session, and lifecycle
  • Human handoff, lead capture, and callback
Voice AISIPMulti-tenantMCPWebRTC
Read the blog↗
Multi-agent · RAG02

Athena

Autonomous competitive intelligence

Athena preview

One LangGraph StateGraph that researches a company end-to-end: a planner you approve, parallel scrapers, a runtime ChromaDB index for RAG, and a critic that loops the analysts until the work is good enough.

LangGraphLangChainFastAPIChromaDBNext.js
GitHub↗
Education · Tooling03

Algolume

The interactive DSA handbook

Algolume preview

A local-first DSA handbook. Don't just read about an algorithm, watch it run: step through real executions, rewind any line, and run Python in-browser via Pyodide.

ReactViteTypeScriptPyodideSupabase
Live site↗
Thesis · Research04

PNEUMA

Long-term video memory architecture

My B.Sc. thesis: a model-agnostic long-term video memory for egocentric QA. Three tiers (SQL, a vector DB, and a knowledge graph) with a cost-based planner that picks the fastest path per query.

  • 72.8% zero-shot on NextQA
  • 98% recall on temporal queries
SQLVector DBKnowledge GraphsQuery Planning
PNEUMA preview
More from the workshop

DynApex

Systems

A C++/OpenMP tool that analyzes dependencies, profiles hot paths, and dispatches work across CPU and GPU execution modes.

GitHub↗

RAG Medical Chatbot

RAG

Chat with medical PDFs. Hugging Face embeddings into a vector store, retrieval-augmented answers, and a Flask UI with session state.

GitHub↗

Realtime Chat

Realtime

Realtime chat with bidirectional sync, typing indicators, and MongoDB-backed sessions on Socket.IO.

GitHub↗

Fitness App

Mobile

Cross-platform fitness app with auth, workout tracking, progress charts, and fitness-API integration.

GitHub↗
02Experience

Where I've done the work

Research R&D, production SaaS, and a bookstore I scaled into a real business.

Product Engineer

Obby LLC · USA (Remote)

May 2025 – Present

Shipping B2B SaaS and production agentic systems.

Voice AI
Production AI voice agents and Agentic Council frameworks for multi-tenant SaaS.
Agentic AI
Multi-agent workflows with tool orchestration, MCP automation, and decision loops.
Growth
Acquisition tracking, inbound lead flows, and automations to scale operations.
Voice-agent architecture blog↗

Research Assistant · AI/ML Engineer

NSU MI Lab × RobotBulls · Dhaka & Luxembourg (Hybrid)

Apr 2025 – Nov 2025

Led R&D on real-time, audio-driven talking-avatar systems.

Talking Avatars
Real-time audio-driven facial animation: lip-sync, gestures, and customizable avatars.
Generative Lip-Sync
Diffusion- and GAN-based video pipelines with idle motion and realism metrics.
Low-Latency Serving
GPU inference with multi-GPU orchestration and WebRTC/RTMP streaming.
Demo↗

Software Engineering Intern

AGT Software Partners · Dhaka & USA (Remote)

Aug 2024 – Jun 2025

Built a secure employee platform with AI analytics on top.

Employee Platform
Secure standups, project, and time-tracking app with Next.js, TypeScript, Socket.IO.
AI & Analytics
Hugging Face chatbot, LLM summaries, ticket categorization, dashboards, and exports.
Security
Token revocation, rate limiting, and session management across the platform.

Founder & Operator

MK Bookhouse · Bangladesh · Remote

Online bookstore

An online bookstore I launched and scaled to profitability with minimal startup capital.

Scale
Sold 50,000+ books and served 10,000+ customers across multiple channels.
Operations
Ran the business remotely with efficient workflows and customer-service systems.
Business
Hands-on with team leadership, supply chain, marketing, sales, and financial planning.
MK Bookhouse on Facebook↗
03Toolkit

What's in the kit

Grouped by where it lives in a system.

01

LLMs & Agents

LangChainLangGraphRAGChromaDBMCPMulti-agentAgentic Council
02

AI / ML

PyTorchTensorFlowDiffusionGANsKnowledge GraphsVector DBsOpenCV
03

Backend

FastAPIREST APIsNode.jsSocket.IOMERNSQL
04

Realtime & Media

WebRTCRTMPNVIDIA TritonCUDAFFmpegLow-latency serving
05

Frontend

Next.jsReactTypeScriptTailwindViteReduxZustand
06

Languages & Tools

PythonC++CJavaJavaScriptGitn8nPyodide
04Leadership & Teaching

Passing it on

Teaching and contest work at North South University.

Computer Science & Engineering01

Undergraduate Teaching Assistant

North South University

  • Ran lab sessions, office hours, and grading for CSE coursework.
  • Helped students debug assignments and reason through core concepts.
Competitive Programming02

Instructor · Problem Solvers Bootcamp

North South University

  • Taught algorithms, data structures, and contest strategy.
  • Mentored students through problem walkthroughs and live contests.
Programming Contests03

Problem Setter · Intra-NSU Contests

North South University

  • Designed algorithmic problems for two intra-university contests.
  • Authored statements, constraints, test cases, and editorials.
05About

A bit about me

I'm an AI engineer who likes the whole stack of a system: the agent graph, the retrieval layer, the GPU serving, and the interface someone actually touches.

At Obby I build multi-tenant voice agents and agentic workflows; before that I led R&D on real-time talking-avatar pipelines. Competitive programming taught me to respect the algorithm under the hype, so I'd rather ship a system that holds up than a demo that photographs well.

Education

North South University

Sept 2021 – Dec 2025

B.Sc. in Computer Science & Engineering · CGPA 3.86 · Dhaka, Bangladesh

Competitive Programming✦
Codeforces Expert↗

Max rating 1650 · @mahir1

ICPC / IUPC highlights

  • 18th / 130AUST IUPC 2025NSU_PyramicSchemers
  • 20th / 100CUET IUPC 2024NSU_MinimumStonkingTriple
  • 29th / 196JU NCPC 2024NSU_OneLastTime
  • 53rd / 313ICPC Asia Dhaka Regional 2025NSU_EktaNaamDaoFast
  • 67th / 300ICPC Asia Dhaka Regional 2024NSU_PyramicSchemers
06Contact

Got a system that needs building?

I’m open to AI engineering roles and interesting collaborations. The fastest way to reach me is email, and I usually reply within a day.

✉Email03mahirshahriar@gmail.com→⌘GitHubgithub.com/mahirshahriar1→inLinkedInin/mahir-shahriar-tamim→★Codeforces@mahir1 · Expert→↓RésuméDownload PDF→