Open to opportunities

Sheharyar Akhtar

@sheharyarakhtar

Message

I build LLM evaluation infrastructure and agentic QA systems for enterprise AI products.

Pakistan

Message

At Turing, I co-built a multi-tenant Auto-Reviewer Evaluation Platform serving LLM QA agents across seven client projects, including Microsoft and Alibaba. I design quality dimensions, build agentic QA flows, and deliver multi-model judging alongside 100+ deterministic checks.

I also founded Markov Chain Finance, where I built live quantitative research tools for regime detection, portfolio optimization, and LLM-powered analysis. Earlier, I delivered RAG and multi-agent systems, real-time ML infrastructure, and Bayesian call-routing models that improved RPC for telecommunications clients.

Experience

Work history, roles, and key accomplishments

Current

Finance Domain Expert

Current

micro1 - Project Navara

Jan 2026 - Present (8 months)

Author complex office-document tasks for AI training data collection and evaluate model responses across quality dimensions for preference data, specializing in DCF/LBO models, investment memos, and portfolio analysis.

Excel Word PowerPoint DCF LBO Investment Memos Portfolio Analysis AI Training Data

Current

Founder & Principal Researcher

Current

Markov Chain Finance

Aug 2025 - Present (1 year 1 month)

Built and deployed a full-stack quantitative research platform with three live tools, including Phase Space Regime Analyzer and Portfolio Optimizer. Developed phase space regime detection with Markov transition matrices and validated on 3,653 days of BTC-USD data.

React Flask Azure OpenAI Markov Models Survival Analysis Quantitative Research Manim

Current

AI Engineer | LLM Evaluation Infrastructure Lead

Current

Turing

Sep 2024 - Present (2 years)

Co-built Turing's central Auto-Reviewer (AR) Evaluation Platform, a multi-tenant backend serving LLM QA agents to 7 client projects. Owned the full client onboarding cycle and built standalone task audit, repair, and evaluation pipelines for ServiceNow's EnterpriseOps-Gym RL benchmark.

LLM As Judge Multi Model Judging AI Agents QA Python SQL GCP Docker Claude Gemini GPT Lang Chain LangGraph Crew AI RAG RL Benchmark Trajectory Analysis

Consultant - Data Science

Systems Limited

Aug 2023 - Nov 2024 (1 year 3 months)

Created RAG and multi-agent workflows using LangChain, LangGraph, CrewAI, Azure OpenAI, and Bedrock APIs, including a doctor-patient LLM scheduling agent reducing manual coordination ~40%. Architected enterprise AI/ML solutions across 20+ proposals contributing $5M+ to pipeline.

RAG Lang Chain LangGraph Crew AI Azure OpenAI Bedrock Kafka AWS Kinesis PySpark MLFlow Python SQL

Data Scientist II - Applied AI

Afiniti

Apr 2021 - Aug 2023 (2 years 4 months)

Built Bayesian Item Response Theory models in R and CmdStan for real-time call routing optimization, delivering +8% RPC for Telefonica Spain, +5% for Masmovil, and +10% for Telecom Italia. Automated gain forecasting via LightGBM sustaining 85%+ accuracy.

Bayesian Inference (PyMC)R Stan LightGBM Apache Superset Drift Monitoring Alerting