My research is centered on how language and retrieval systems model people: how they infer preferences from behavior and dialogue, and how reliable those user representations remain across interactions. I am particularly interested in the intersection of large language models, personalization, conversational systems, retrieval, and reliability.
Before focusing more deeply on research, I spent over four and a half years at
NeshanMaps
building search, geocoding, and ML-assisted systems at production scale. That experience strongly shapes how I think about evaluation: a model is useful only when its behavior can be measured against real users and real constraints.
Latest Updates
Sep 2026
Revising AuDiPE for its next submission.
Sep 2026
Developing my MSc thesis into a research manuscript on personalization in conversational recommender systems.
Mar 2026
Wrapped up over four and a half years at NeshanMaps, working across search, ML, geocoding, and LLM-assisted systems.
Current Work
MSc Thesis
Ongoing
2025–2026
Personalization & Diversification in Conversational Recommender Systems.
The thesis studies how conversational recommenders can model user preferences from dialogue while balancing recommendations aligned with known preferences and maintaining diversity over the course of a conversation.
PersonalizationInformation RetrievalConversational Systems
Experience
NeshanMaps
Over four and a half years across search, ML, backend systems, and product engineering.
Aug 2021 – Mar 2026
NeshanMaps · Gamification Team
Software Engineer
Mar 2025 – Mar 2026
Integrated LLMs into comment moderation, reducing manual review effort by roughly 60%.
Automated validation workflows and improved production reliability for gamification systems.
NeshanMaps · Search Team
Engineering Manager
Jul 2024 – Mar 2025 (concurrent)
Led a team of four engineers and owned backend design for search-related systems.
Worked on ML-driven ad placement and suggestion systems, contributing to a roughly 2× increase in ad revenue.
NeshanMaps · Search Team
Software Engineer
Aug 2021 – Mar 2025
Improved search CTR from 57% to 64% through search improvements and experimentation.
Redesigned geocoding to handle roughly 3× more load, reduce latency by about 15%, and improve accuracy by about 20%.
Worked on search infrastructure serving a product that grew from roughly 10M to 30M+ users.
Publications
In Revision
AuDiPE: Behavior-Grounded User Simulation via Audited Dialogue-based Profile Enrichment for Conversational Recommender Systems
A. Roshandel*, E. Shahabi*, M. Sadeghi, A. Shakery · *equal contribution
2026
AuDiPE studies reliable user simulation and profile enrichment for conversational recommender systems, with a focus on evaluating the quality and robustness of inferred user profiles.
Away from research and engineering, I enjoy films and TV series, long walks, listening to music,
traveling, reading novels, and charcoal drawing. I like activities that give me space to slow down,
observe details, and come back to work with a clearer head.
Films & TVWalkingMusicTravelNovelsCharcoal Drawing
Contact
I'm looking for PhD opportunities starting in 2027 where I can work on large language models—especially how they model users, personalize their behavior, and remain reliable.
