Berkeley, CA · open to opportunities

Parham builds systems for autonomy

EECS & MEng at UC Berkeley. Research on safe imitation learning, multi-agent coordination, and LLM reasoning.

↓ Resume Projects Publications About GitHub ↗
Portrait of Parham Sharafoleslami
Now Decentralized fleet coordination for airport ground ops · MEng thesis on safe imitation learning for race-car controllers

Selected work

all 23 →
Aerospace · 2026

An Autonomous Fleet Manager for 1,600 Satellites

Guidance, navigation, and control software that runs a constellation of electric propulsion satellites on its own. It keeps every satellite in its place in the formation, works out which close passes are real threats, and moves a satellite out of the way only when it has to. Designed around an eight mode state machine, tested in a simulated world full of errors the software is never told about, at the size of a full Starlink shell, and checked against CelesTrak's SOCRATES on the real satellite catalog.

PythonC++17pybind11NumPySciPySGP4
Read case study →
research
Machine Learning · 10 min read

Does a Merge Planner Need a Model of the Traffic?

Three planners share one control interface and differ only in what they believe about the car behind them: nothing, everything, or whatever they can learn. Across 41,130 episodes on 914 held-out Waymo scenes, a correct model of the follower is worth everything in one narrow band of gap sizes and nothing outside it, and the moment it is wrong, the planner that trusted it drops to zero. A learned policy wins the opposite regime, but a control proves it wins by acting early rather than by understanding anything.

PythonJAXWaymaxPPO +2
Robotics · 20 min read

Nonlinear MPC for Autonomous Racing

A controller for a full-scale Dallara AV-24 in the Indy Autonomous Challenge, built on two timescales: a minimum-time lap solved offline with a double-track Pacejka tyre model, and a progress-maximizing MPC that drives it at 100 Hz. My part is the offline layer, which produces the racing lines the online controller follows. Validated in software in the loop, in simulation at Laguna Seca (117.6 s, 25% faster than the baseline controller), and on the physical car.

PythonC++17CasADiacados +4
Robotics · 7 min read

MPC for UR7e Robotic Arm: Warehouse Sorting

Constrained joint-space Model Predictive Control for a UR7e manipulator. An RGB-D perception pipeline localizes objects and obstacles; a receding-horizon CasADi/IPOPT solver generates collision-free joint trajectories at 12 Hz, validated in MuJoCo and deployed on real hardware.

PythonMuJoCoCasADiROS2 +1
Robotics · 8 min read

Decentralized Fleet Coordination for Airport Ground Operations

Multi-agent coverage control for airport ground vehicles using Buffered Voronoi Cells, Lloyd's algorithm, and game-theoretic demand response. Zero collisions across all test scenarios.

PythonMulti-Agent SystemsOptimal ControlGame Theory
Machine Learning · 3 min read

Deep Learning from Scratch

The full arc: backpropagation in raw NumPy, CNNs with BatchNorm and Dropout trained to 78% on CIFAR-10, multi-head self-attention for text summarization, and a Masked Autoencoder that reconstructs images from 25% of their patches, then transfers those features to downstream tasks.

PythonNumPyPyTorchCIFAR-10
Machine Learning · 4 min read

Classical ML from Scratch

Two learning paradigms built from NumPy up: tree-based spam classification (decision tree, Random Forest, AdaBoost) and SVD/ALS matrix factorization for movie recommendations. No frameworks; matched scikit-learn on both.

PythonNumPySciPyMachine Learning

Selected publications

all →
Preprint · 2026 · advisor

Visualizing and Benchmarking LLM Factual Hallucination Tendencies via Internal State Analysis and Clustering

Nathan Mao , Varun Kaushik , Shreya Shivkumar , Parham Sharafoleslami , Kevin Zhu , Sunishchal Dev

FalseCite, a curated dataset of 82k false claims paired with fabricated citations, reveals that LLMs hallucinate more readily when misleading references are present, especially in smaller models like GPT-4o-mini. Hidden-state clustering exposes a distinctive 'horn-like' geometry across hallucinating and non-hallucinating activations.

arXiv ↗ PDF ↗ #hallucination#benchmark#interpretability
Preprint · 2025

COMPASS: Context-Modulated PID Attention Steering System for Hallucination Mitigation

Kenji Sahay , Snigdha Pandya , Rohan Nagale , Anna Lin , Shikhar Shiromani , Parham Sharaf , Kevin Zhu , Sunishchal Dev

A decoding-time intervention that dynamically steers attention toward retrieved context using a PID controller driven by a per-head Context Reliance Score. No retraining, no multi-pass decoding: just interpretable, single-stream control of evidence grounding. Reduces hallucinations by 2.8–5.8% absolute across HotpotQA, XSum, HaluEval, and RAGTruth.

arXiv ↗ PDF ↗ #hallucination#decoding#attention#interpretability
NeurIPS 2025 · 2025 · advisor

Optimizing Chain-of-Thought Confidence via Topological and Dirichlet Risk Analysis

Abhishek More , Anthony Zhang , Nicole Bonilla , Ashvik Vivekan , Kevin Zhu , Parham Sharafoleslami , Maheep Chaudhary

EDTR estimates LLM confidence by treating each chain-of-thought as a vector in semantic space and analyzing the geometry of the reasoning distribution. Combined with Dirichlet-based uncertainty quantification, it achieves 41% better calibration than competing methods and perfect accuracy on AIME.

arXiv ↗ PDF ↗ #calibration#chain-of-thought#uncertainty

Working with

Python 18 NumPy 10 PyTorch 4 Computer Vision 4 SciPy 3 C++17 2 pybind11 2 CasADi 2 ROS2 2 Optimal Control 2 Game Theory 2 scikit-image 2 RISC-V 2 Information Theory 2 Matplotlib 2 Java 2