What does Mehran Dashti do?

I am a senior software engineer specialising in multi-agent AI systems and high-scale distributed backends. Day to day that means designing LangGraph agent orchestration, building retrieval-augmented generation pipelines on Milvus, hardening LLM systems against prompt injection and PII leakage, and decomposing monoliths into event-driven services in Python and FastAPI. I have been doing backend work for over ten years and AI platform work for the last two.

Are you available for hire?

Yes. I am open to senior backend, staff engineer and AI engineering roles, and to consulting engagements. I work remotely from Tehran with teams in Europe and worldwide, and I currently work remotely for a German company, so distributed collaboration across time zones is normal for me rather than an experiment.

What kind of engagements do you take?

Three shapes. Full-time senior or staff engineering roles. Fractional or part-time work on an AI platform, typically one to three days a week. And short scoped consulting: an architecture review of an agent system, a RAG pipeline that is returning poor results, an AI security assessment covering prompt injection and PII handling, or a monolith decomposition plan. Scoped work usually runs two to six weeks.

Can you help with AI security and data protection?

That is one of my core areas. I have implemented prompt-injection defense in production — fencing, closed-schema extraction, hard-block chokepoints and audit logging — and designed a reversible PII masking and pseudonymization layer with Microsoft Presidio, spaCy NER and custom recognizers, applied across agent pipelines with tenant isolation and fail-closed controls.

What is your experience with RAG and vector search?

I build and tune retrieval systems on Milvus: BGE-M3 embeddings, hybrid semantic and lexical retrieval, reciprocal-rank fusion, weighted multi-signal reranking and relevance gating, with offline evaluation using MRR and precision@k. I also built the embedding and ingestion service that keeps such a pipeline current. There is a working reference implementation in my fastapi-milvus-sample repository if you want to read the code before you talk to me.

Which languages do you work in?

Persian is my first language, I work professionally in English, and I have working German. Technically I write Python and TypeScript day to day, with PHP and Laravel from earlier years, and I have shipped multilingual agent systems covering German, English, Persian, Arabic and Turkish.

How do I get in touch?

Email mehran19.md@gmail.com with a paragraph on the problem and the constraints. I reply to concrete enquiries within a day or two. If it is a fit I will suggest a call; if it is not, I will say so and point you somewhere better.

Mehran Dashti — Senior AI & Distributed Systems Engineer

whoami

Profile

Senior Software Engineer with 10+ years delivering high-scale distributed systems across fintech, e-commerce, logistics, and AI.

Led decomposition of a large production monolith into 10+ independently deployable microservices at LiaServices. Built Iran's first Agent-as-a-Service platform at Delaverse (Jan 2025 – Apr 2026). Deep expertise in event-driven architecture, LLM integration, and high-throughput API design. Currently seeking senior backend or staff engineer roles at high-scale product companies in Tehran or remote.

10+

Microservices shipped

99.9%

Payment uptime

10k

Requests / min peak

60k+

Users served

systemctl status agent-pipeline

Agent Systems

How the Agent-as-a-Service platform routes a request.

01
Intent
Versioned REST entry, auth & tenant resolution.
02
Orchestrator
Plans the run, selects agents, splits the workflow.
03
Tools
Plugin system — retrieval, HTTP, SQL, queue workers.
04
Memory
Redis-backed state and conversation context per tenant.
05
Response
Streamed output, traced and monitored in real time.

Multi-agent

Orchestration engine with plugin system and workflow pipeline.

LLM integration

OpenAI API in production — Persian language support end to end.

Observability

Real-time monitoring dashboard, sub-second agent response times.

git log --graph --all

Experience

Ten years of systems in production.

4 roles · 20 systems
2016 → present

6062c21(HEAD -> main, origin/tunoo)

langgraph · multi-agent

Now

Tunoo

AI Software Engineer
Germany · Remote
Present

active

LangGraphRAG · MilvusPII safety

Tunoo AI — Multi-Agent AI Platform. Architected and built a production-grade multi-agent AI platform using LangGraph for orchestrating complex agent workflows in Python.

75bc76fAgent Orchestration
Designed the agent orchestration flow, managing control and data flow between agents to improve reliability and reduce failure cascades.
Python · LangGraph
16b1626Prompt Injection Defense
Implemented prompt injection defense, detecting and rejecting malicious or adversarial input before it reached downstream agents.
Python · LangGraph
273156dVector Search
Improved Milvus vector search performance, tuning indexing and retrieval configuration for faster, more accurate similarity search.
Milvus · Python
8eb3e80Embedding Service
Built an embedding service to populate and maintain the RAG pipeline, handling document ingestion and vectorization into Milvus.
Python · Milvus · Docker
a8de2b3PII Masking Layer
Designed and implemented a PII masking/anonymization layer using Microsoft Presidio with pseudonymization techniques to protect sensitive data across agent pipelines.
Presidio · Python
568c5c6Spec-driven Development
Established a spec-driven development (SDD) workflow using OpenSpec, improving requirement traceability and reducing implementation drift across features.
OpenSpec

Tech stack: Python, LangGraph, Presidio, Milvus, MongoDB/Redis (as applicable), Docker

b71ccb0(tag: v2020.25)

mesh · 10+ services

2020–25

LiaServices

Senior Software Engineer
Full-time
Jan 2020 – Apr 2025

archived

7 servicesevent-driven99.9% uptime

Core engineer in a 14-person team — led decomposition of a large production monolith into 10+ independently deployable microservices, improving release cadence from monthly to weekly with zero-downtime migrations.

d7e9360Payment Service
Multi-gateway failover with idempotent transactions & automated reconciliation — 99.9%+ uptime; zero duplicate charges across all transaction flows.
Python · MongoDB · Docker
4c1b165API Gateway
Internal reverse proxy routing 100% of cross-service traffic — enforced rate limiting, auth, and service discovery; reduced inter-service latency by eliminating redundant auth round-trips.
Laravel · Redis · Docker
466307aAAA Service
Centralised auth & accounting with LDAP integration and Redis-backed sessions — single source of truth for all 10+ services; session lookup under 5ms via Redis caching.
Laravel · LDAP · MariaDB · Redis
856a767Warehouse Automation
Event-driven inventory & order fulfilment processing 5,000+ orders/day at peak via RabbitMQ — decoupled producers from consumers, eliminating bottlenecks under sustained load.
Laravel · MongoDB · RabbitMQ · Redis · S3
ade51fcLMS
Full learning platform serving part of LiaServices' 60,000+ user base — certifications, video courses & class reservations; S3-backed media delivery kept page load times under 2s for course content.
Laravel · MongoDB · MariaDB · Redis · S3
dce8ed1Ticketing
SLA-aware smart routing engine — priority-based agent assignment cut average first-response time and ensured high-priority tickets were never left unassigned.
Python · MongoDB · Celery · Redis
9657e16Product & Promotion
Flexible catalogue handling 10,000 requests/minute at peak with an order placed every 3 seconds — RabbitMQ-backed async processing prevented promotion bursts from blocking the main request path.
Python · MongoDB · RabbitMQ

-- NORMAL --window 1/4[1-4] ←→ switch

tree -L 1 ~/.skills

Technical Skills

The stack behind the agents.

6 domains
python · langgraph · milvus

~/.skills

01languages-runtimes/
Languages & RuntimesPython 3.11+ (async-first), TypeScript, SQL, Bash
02ai-llm-engineering/
AI / LLM EngineeringLangGraph, LangChain, OpenAI API, structured outputs & schema-validated generation, multi-agent supervisor/worker orchestration, human-in-the-loop interrupt/resume, checkpointed conversational state, prompt engineering, intent routing & disambiguation, multilingual agents (DE/EN/FA/AR/TR)
03rag-retrieval/
RAG & RetrievalMilvus, BGE-M3 embeddings, hybrid semantic + lexical retrieval, reciprocal-rank fusion, weighted multi-signal reranking, relevance gating, corpus ingestion pipelines, offline retrieval evaluation (MRR, precision@k)
04ai-security/
AI SecurityPrompt-injection defense (fencing, closed-schema extraction, hard-block chokepoints, audit logging), reversible PII masking/pseudonymization (Presidio, spaCy NER, custom recognizers), GDPR-oriented data handling, least-privilege & tenant isolation, fail-closed controls, staged rollout with kill-switch
05backend-distributed-systems/
Backend & Distributed SystemsFastAPI, WebSocket, Kafka (event-driven architecture), Redis (leases, dedup, registry), PostgreSQL, MongoDB, S3/MinIO, idempotent operations, graceful degradation, fault isolation
06platform-practice/
Platform & PracticeDocker, Kubernetes/OpenShift, GitLab CI/CD, SonarQube quality gates, OpenTelemetry/Langfuse tracing, structured logging, pytest (unit/contract/integration), spec-driven development, architectural documentation
07education.txt
EducationB.Sc. Software Engineering — Shamsipour University · TehranOct 2015 – Oct 2019

6 directories, 1 file

gh repo list --source

Open Source

Open Source
name	description	lang	license
fastapi-milvus-sample	Retrieval-augmented generation service on FastAPI, Milvus and OpenAI — standard, conversational and LangGraph agentic RAG, with five deep-dive guides to the stack.	Python	—
fastapi-iranian-bank-gateways	FastAPI toolkit for all 17 major Iranian payment gateways — async-first, production-ready.	Python	MIT
fastapi-role-permission	Role & permission middleware for FastAPI with JWT auth and multi-tenant support.	Python	MIT
fastapi-blueprint	Production-ready FastAPI scaffold — DI, DB, auth, Docker out of the box.	Python	—
laravel-api-response	Standardised JSON response wrapper for Laravel APIs.	PHP	—
laravel-api-exception	Structured JSON error handling for Laravel — consistent HTTP status codes.	PHP	—

6 repositories · github.com/MehranDashti →

mehran-agent --topic hire

Hire

What I do, and how to start.

Short answers to the questions people actually ask before getting in touch.

What does Mehran Dashti do?: I am a senior software engineer specialising in multi-agent AI systems and high-scale distributed backends. Day to day that means designing LangGraph agent orchestration, building retrieval-augmented generation pipelines on Milvus, hardening LLM systems against prompt injection and PII leakage, and decomposing monoliths into event-driven services in Python and FastAPI. I have been doing backend work for over ten years and AI platform work for the last two.
Are you available for hire?: Yes. I am open to senior backend, staff engineer and AI engineering roles, and to consulting engagements. I work remotely from Tehran with teams in Europe and worldwide, and I currently work remotely for a German company, so distributed collaboration across time zones is normal for me rather than an experiment.
What kind of engagements do you take?: Three shapes. Full-time senior or staff engineering roles. Fractional or part-time work on an AI platform, typically one to three days a week. And short scoped consulting: an architecture review of an agent system, a RAG pipeline that is returning poor results, an AI security assessment covering prompt injection and PII handling, or a monolith decomposition plan. Scoped work usually runs two to six weeks.
Can you help with AI security and data protection?: That is one of my core areas. I have implemented prompt-injection defense in production — fencing, closed-schema extraction, hard-block chokepoints and audit logging — and designed a reversible PII masking and pseudonymization layer with Microsoft Presidio, spaCy NER and custom recognizers, applied across agent pipelines with tenant isolation and fail-closed controls.
What is your experience with RAG and vector search?: I build and tune retrieval systems on Milvus: BGE-M3 embeddings, hybrid semantic and lexical retrieval, reciprocal-rank fusion, weighted multi-signal reranking and relevance gating, with offline evaluation using MRR and precision@k. I also built the embedding and ingestion service that keeps such a pipeline current. There is a working reference implementation in my fastapi-milvus-sample repository if you want to read the code before you talk to me.
Which languages do you work in?: Persian is my first language, I work professionally in English, and I have working German. Technically I write Python and TypeScript day to day, with PHP and Laravel from earlier years, and I have shipped multilingual agent systems covering German, English, Persian, Arabic and Turkish.
How do I get in touch?: Email mehran19.md@gmail.com with a paragraph on the problem and the constraints. I reply to concrete enquiries within a day or two. If it is a fit I will suggest a call; if it is not, I will say so and point you somewhere better.

bash

Shell

Ask it anything the CV knows.

Mehran Dashti — interactive shell

locale en · type a command, or tap one below

mehran@dashti:~$ whoami

Mehran Dashti

Senior Software Engineer — AI & Distributed Systems

AI Software Engineer @ Tunoo — Germany · Remote

10+ years · fintech, e-commerce, logistics, AI

type 'help' for the full command list

mail -s "hello" mehran

Let's build
something durable.

Open to senior backend and staff engineer roles — Tehran or fully remote.

Emailmehran19.md@gmail.com Phone+98 919 720 8607 Phone · 2+98 939 055 1986 GitHubMehranDashti LinkedInin/mehran-dashti