Raunit Kohli — Consolidated Reference Notes

Purpose: Single reference file for all raw content, research, links, and context about Raunit’s work, projects, and career goals. Intended for use by AI agents and for personal reference when updating the resume, CV, or website. Last updated: April 2026.

Table of Contents

1. Career Goals & Target Roles
2. Job Market Research (April 2026)
3. Graduate School Research
4. Rocket Software — Detailed Content
5. Chiba Lab (RUBI Lab) — Detailed Content
6. Publications
7. Projects
8. Technical Stack & Skills
9. Website & Resume Architecture
10. Key Decisions & Distinctions
11. Links & References

Career Goals & Target Roles

Raunit is targeting mid-career software engineering roles focused on:

• Reinforcement Learning (policy optimization, sim-to-real transfer)
• Robotics Automation (autonomous behavior, multi-agent systems, ROS2)
• Agentic AI Applications (LLM-powered agents, multi-agent workflows, tool use)

Also interested in graduate school (MS or PhD) in robotics/RL/AI at top programs.

Job Market Research (April 2026)

Industry Roles Researched

Searched for 2026 postings at companies aligned with Raunit’s interests:

Figure AI — Robotics ML Engineer

• Keywords: PyTorch, RL, sim-to-real, policy optimization, C++, real-time systems
• Looking for: experience with physical robots, manipulation, locomotion

Skild AI — ML Engineer, Foundation Models for Robotics

• Keywords: large-scale RL training, simulation (MuJoCo/Isaac), multi-task learning
• Looking for: sim-to-real transfer, generalizable robot policies

NVIDIA — Robotics Research Scientist (Isaac Lab)

• Keywords: Isaac Sim/Gym, RL, sim-to-real, GPU-accelerated simulation, C++/CUDA
• Looking for: publications, simulation expertise, real robot deployment

Apple — ML Engineer, Robotics & Automation

• Keywords: RL, computer vision, motion planning, sensor fusion, ROS
• Looking for: production ML systems, real-world deployment experience

General patterns across all postings:

• PyTorch is universal for robotics/RL
• C++ is highly valued (real-time constraints)
• ROS/ROS2 is table stakes for robotics roles
• Sim-to-real transfer is a hot keyword (MuJoCo, Isaac, Habitat)
• Publications are a plus but not always required for industry
• Multi-agent systems / swarm robotics gaining interest
• LangChain/agentic AI is its own fast-growing category separate from robotics

Gap Analysis (Profile vs. Market)

Strengths already present:

• PyTorch RL implementation with real robots (PiRat) — very rare and valuable
• ROS2 experience with actual hardware deployment
• C++ production CV work (300% speed improvement)
• Multi-agent experimental data (2500+ sessions)
• Published research (3 papers, 2nd author on ICDL)
• Agentic AI production work (LangChain, DSPy, ReAct agents)
• Leadership (1800+ engineers trained, AI Center of Excellence)

Gaps identified and addressed:

• ✅ “Reinforcement Learning” keyword was missing from skills → added to Domains
• ✅ Libraries line was commented out in LaTeX → restored with DSPy, MLflow added
• ✅ ROS2 not explicit (was just “ROS”) → corrected to ROS2
• ✅ MuJoCo/sim-to-real language missing → added via Regulation project
• ✅ No Docker/deployment keywords → added via EVA product
• ⚠️ No public GitHub repos (most code NDA’d/under grants) — user may create 1-2
• ✅ Google Scholar profile found and added: https://scholar.google.com/citations?user=abZZ14QAAAAJ
• ✅ PiRat RL project page on website filled with content (was placeholder)

Graduate School Research

Programs Researched

• CMU Robotics Institute (MS/PhD): Strong RL + robotics focus. Values publications, real robot experience, C++ systems work.
• Stanford CS (MS/PhD AI track): Values breadth across ML + systems. Publications important for PhD.
• MIT CSAIL (Robotics): Hardware + software integration valued. Real-world deployment experience a plus.
• UC Berkeley BAIR (PhD): Strong RL tradition (Abbeel, Levine). Sim-to-real transfer research.

What Grad Schools Want (CV Optimization)

• Publications are critical (Raunit has 3, including 2nd author ICDL 2025)
• Research experience duration matters (Raunit: 4+ years)
• Recommendation letters from PIs (Chiba Lab relationship is strong)
• Technical depth > breadth on CV
• Teaching experience is a plus (TA + AI enablement + VEX coaching)
• GPA threshold usually 3.5+ (Raunit: 3.91 major GPA — excellent)

Rocket Software — Detailed Content

Rocket EVA (AI Platform)

• Product page: https://www.rocketsoftware.com/en-us/products/eva (was 403’d when checked April 2026)
• Raunit owns multiple core features:
  • RESTful API design
  • End-to-end OIDC Authorization pipeline
  • Live telemetry/streaming pipelines
  • Asynchronous tool execution pipeline
  • Fixed at least half the coding bugs for first product release
  • Structured Docker deployment pipeline
  • Owning integration with Rocket teams’ products into EVA
  • Worked with ID teams for internal and public-facing documentation
• Most aggressively promoted and marketed company-wide tool

SAS-to-Python Code Translation

• Specialized team project using multiple LLM-optimization techniques
• Built large-scale datasets and model evaluation pipelines
• Used ReAct agent (Reasoning+Action agent) via LangChain
• Collaborated on Agentic Context Engineering (ACE) optimizer
  • ACE paper: https://arxiv.org/abs/2510.04618 (ICLR 2026)
  • IMPORTANT: Raunit is NOT an author on this paper — he collaborated on/used the framework
• Created multi-agent workflows for: planning, architecture design, test generation, custom live code execution, refinement loops
• Integrated with MLflow for logging/evaluation

SmartChat (RAG System)

• Brochure: https://www.rocketsoftware.com/sites/default/files/resource_files/smart-chat-brochure.pdf
• Raunit built: SQL generation/execution pipeline, Citations feature (multi-Gaussian clustering, 80% improvement), Feedback API, Reranking feature, AWS CloudWatch logging system
• NL-to-SQL retrieval agent with regex-based syntax validation
• Fine-tuned transformer-based sentence embedding models with custom tokenizers

AI Enablement & Teaching

• Created entire AI foundations and enablement lecture series for NextGen program
  • NextGen Academy: https://www.rocketsoftware.com/en-us/nextgen-academy
• Delivered AI foundations lecture series at global Rocket offices, achieving 100% content application rate
• 100% content application rate (every audience member applied tools/techniques to daily work)
• Workshops solve real-world problems on existing Rocket engineering teams
• Leading collaboration with 3rd party vendor for company-wide AI enablement (1800+ engineers)
• Week-long lecture series covering: GenAI foundations, LLM conceptual understanding, prompt engineering, model tuning, workflow building, AI-assisted coding

Smart Scrum Master (Hackathon)

• 2025 Rocket.Build Hackathon — Won “Most Innovative” award (1 of 6 winners from 650+ projects globally)
• LLM-powered Agile assistant optimizing Jira workflows
• Securely updates/creates/searches/summarizes Jira tickets based on user queries or Teams meeting content
• Uses reasoning agent for intelligent Jira board organization
• Exposed as MCP Server — can be hosted by other chatbots and IDEs
• Project served as a foundation for AI-focused workflow optimizations and developer enablement trainings
• Acted as basis for subsequent AI-focused optimizations and developer enablement trainings

Other Rocket Details

• AI Center of Excellence — working with the CoE to bring advanced AI tooling to expert engineers (NOT a founding member; does NOT collaborate with DSPy creators — DSPy is just a library used)
• Main languages/libraries: Python, DSPy, LangChain, PyTorch, HuggingFace
• MultiValue Experience Unit Test Coverage project (earlier role): launched foundational unit test framework with OOP mocking, 30% line coverage (3x OKR goal)

Chiba Lab (RUBI Lab) — Detailed Content

Lab Info

• Full name: Robotics Using Bayesian Inference Lab, UC San Diego
• PI: Professor Andrea A. Chiba
• Lab website: https://www.chiba-lab.org/ (unfinished as of April 2026)
• Raunit’s tenure: Staff ML Research Engineer (01/2021–09/2024), ML Research Consultant (09/2024–present)

PiRat Robot System (Technical Deep-Dive)

From content.md’s detailed tech stack description:

The Robot:

• Rat-sized mobile robot for studying social behavior in rats
• Electronics: three stacked circuit boards (Raspberry Pi Zero footprint)
  • Top: Raspberry Pi Zero (Raspbian, Wi-Fi via wireless card)
  • Middle: Distribution board with STM32 F042 microcontroller (USB to Pi Zero, PWM to encoders, USART to driver)
  • Bottom: Custom two-gimbal motor driver based on Martinez Gimbal board
• Weight: 0.24kg, top speed: 1.1m/s, top angular velocity: 4.7m/s
• Reference: “PiRat: An autonomous framework for studying social behaviour in rats and robots.” IROS 2018.

Behavior Controller (GUI):

• External GUI built in Python with Kivy
• Captures live global top-down positions of all agents in circular arena
• Calculates target linear/angular velocity based on PiRat position/orientation
• Multiple finite-state machine inspired autonomous behaviors
• Homebase detection: convolution sliding kernel over frequency distribution of positions
• Target position: next pose on trajectory that intersects or circumvents homebase
• Updates Kivy GUI + sends velocity commands via ROS2

Tracking System:

• Multi-threaded Python script
• Live-inference pose estimation using SLEAP (Social LEAP Estimates Animal Poses)
• Pipeline: load pre-trained neural networks → capture RGB frames → inference → dynamic memory array → visualization overlay
• Poses streamed to behavior controller in real-time

NeuroPos Composable Pipeline:

• Location: /Users/raunit/Desktop/ChibaLab/NeuroPosHb/
• Modular processing steps (discovered from directory):
  • HomebaseDetection (convolution kernel)
  • SpectralFeatures
  • SpatialAnalysis
  • Signal processing modules
• Integrated pose-estimation + neural signal data analysis
• Used across 2500+ experimental sessions

PiRat Software Stack:

• Location: /Users/raunit/Desktop/ChibaLab/pirat/
• Multi-threaded SLEAP + Kivy GUI + ROS behavior control
• ROS2 (NOT ROS1)

RL Regulation Framework

• MuJoCo simulation environment
• Uses Google DeepMind tools
• Multiple policy optimization algorithms (user said “multiple policy algorithms” — specific names like PPO/SAC not confirmed for this project; kept general)
• Models exploration-regulation dynamics in autonomous robotic behavior
• Bio-influenced: mimicking rat grooming behaviors for positive exploration-regulation effects
• Ongoing/confidential — keep descriptions general
• Designed for sim-to-real transfer to physical PiRat robot

AI Classroom Project

• Sponsor: (user clarified: project has nothing to do with HEDC — removed from all docs)
• Multi-million dollar project
• Raunit was founding engineer hired
• Built initial pilot system for multimodal live synced secure data streaming/recording in Pre-K classroom
• Main point of contact for technical specifications
• Architecture based on PTPv2 (Precision Time Protocol v2)
• Components confirmed by user:
  • High-precision Basler cameras
  • Custom Arduino firmware for OpenBCI Emotibits (biophysical signal detection)
  • Lab Streaming Layer (LSL)
  • PTPv2 synchronization
• Exhaustive knowledge transfer sessions with new hires
• Note: “Dante networking”, “on-prem quartz crystal grandmaster clock”, “neuromorphic sensors”, and “wearable physiological monitors” were found during directory exploration but NOT confirmed by user — do not include on resume without confirmation
• Can be categorized under either Staff Researcher or Consultant role

Research Team Management

• Hired and trained undergraduate research assistants (RAs)
• RAs learned system operation and experiment execution
• Established knowledge transfer documentation for long-term project continuity

Publications

See Publications section below.

Publications

1. ICDL 2025 (2nd author): I. Jackson, R. Kohli, E. Leonardis, V. R. de Sa, S. Fei, L. Quinn, Y. Lou, A. A. Chiba. “Explore-Exploit Behaviors During Rat-Robot Interactions Optimize Social and Spatial Security.” IEEE International Conference on Development and Learning, 2025.
   • IEEE: https://ieeexplore.ieee.org/document/11204421
2. Frontiers in Psychology 2022: E. J. Leonardis, L. Breston, R. Lucero-Moore, L. Sena, R. Kohli, L. Schuster, L. Barton-Gluzman, L. K. Quinn, J. Wiles, A. A. Chiba. “Interactive neurorobotics: Behavioral and neural dynamics of agent interactions.” Frontiers in Psychology, Sec. Cognitive Science, Vol. 13, 2022.
   • DOI: https://doi.org/10.3389/fpsyg.2022.897603
3. BCI 2025 (Abstract): I. Jackson, R. Kohli, R. Lucero-Moore, Y. Lou, L. K. Quinn, L. Breston, J. Wiles, A. A. Chiba, E. Leonardis. “Robotic Exploratory Control Via Subcortical Oscillations.” 11th International Brain-Computer Interface Meeting, 2025.
   • PDF: https://openlib.tugraz.at/download.php?id=686289f5ddb80&location=browse

Projects

VLM-Guided Hierarchical RL in Habitat-Lab (WPI, 2024)

• Hierarchical PPO/SAC for Fetch robot
• Vision-Language Model high-level planner
• Autonomous navigation and manipulation from language instructions
• Reward shaping and dual-layer hierarchy (removed — these details were AI-hallucinated, not provided by user)

Smart Scrum Master (Rocket.Build 2025)

• See Rocket Software section above

Aspera Workflows Automated Test Suite (IBM, 2023)

• Ruby automation of API endpoint testing for IBM Aspera on Cloud Workflows
• 780+ automated regression tests covering all endpoints
• Identified 15+ critical production bugs
• Full development cycle for API Endpoint Test Controller
• Contributed to global watsonx AI hackathon

VEX Robotics Autonomous Systems Coaching (GHS)

• Mentoring competitive VEX Robotics team
• Autonomous path planning: PID control, cubic spline trajectory generation, odometry-based localization, sensor fusion
• Curriculum covers control theory fundamentals (note: “directly maps to ROS-based architectures” was AI-hallucinated — user did not state this connection)

MultiValue Experience Unit Test Coverage (Rocket)

• Foundational unit test framework with OOP mocking and line coverage
• 30% line coverage (3x initial OKR goal)
• Identified multiple redundant dependencies

Technical Stack & Skills

Languages

Python, C++, Ruby, Java, SQL, Bash

Libraries & Frameworks

PyTorch, DSPy, LangChain, HuggingFace, MLflow, Pandas, NumPy, Scikit-learn, Kivy, OpenBCI

Domains

Reinforcement Learning, Multi-Agent Systems, Agentic AI, NLP, Computer Vision, Sim-to-Real Transfer, Policy Optimization (PPO/SAC), Signal Processing, Pose Estimation, Neurorobotics, Embedded Systems

Tools & Platforms

ROS2, MuJoCo, SLEAP, Habitat-Lab, Docker, AWS (CloudWatch), Kubernetes, Git/GitHub, Atlassian Suite (Jira, Confluence), Arduino, PTPv2, Lab Streaming Layer (LSL)

Coursework

Deep Learning for NLP, Neural Networks, Data Modeling, Advanced Topics in LLMs, Supervised & Unsupervised ML, Data Structures & Algorithms, Markov Decision Processes, Reinforcement Learning (WPI)

Website & Resume Architecture

Website

• Jekyll site on GitHub Pages at raunitkohli.com (CNAME file present)
• Layout: _layouts/default.html → includes header.html + footer.html
• Config: _config.yml with nav items (Home, Industry, Research, Extracurricular, Resume)
• Pages: index.html, work.html, research.html, extracurricular.html, resume.html
• Resume page has dual download buttons (Industry Resume + Academic CV) with PDF preview iframes

Resume Files

• assets/rk_resume.tex — 1-page industry resume (LaTeX source)
• assets/rk_cv.tex — Multi-page academic CV (LaTeX source)
• assets/rk_resume.pdf — Compiled industry resume PDF
• assets/rk_cv.pdf — Compiled academic CV PDF (needs to be generated)
• assets/Raunit_Resume_Feb_25.pdf — Older Feb 2025 version (kept for reference)

LaTeX Template Details

• Based on sb2nov/resume template
• Custom commands: \resumeSubheading, \resumeItem, \resumeItemListStart/End, \resumeSubHeadingListStart/End
• RUBI Lab section uses manual \vspace{-2pt}\item + \begin{tabular*} blocks (two sub-roles under one org)
• Both fontawesome (v4) and fontawesome5 loaded; v4 takes precedence
• Industry resume: aggressive margins (oddsidemargin -0.6in, topmargin -0.7in), negative vspace (-14pt to -20pt)
• CV: slightly relaxed margins (oddsidemargin -0.5in, topmargin -0.5in)

Compilation

cd assets
pdflatex rk_resume.tex   # Industry resume
pdflatex rk_cv.tex        # Academic CV
rm -f *.aux *.log *.out   # Clean artifacts

Requires LaTeX distribution (macOS: brew install --cask basictex)

Key Decisions & Distinctions

• ACE framework (arxiv:2510.04618): Raunit collaborated on/used the ACE framework but is NOT an author. Resume says “collaborating on ACE optimization framework”, not “co-authored.”
• AI Center of Excellence: Raunit works WITH the CoE. He is NOT a “founding member” and did NOT “collaborate with DSPy library creators.” DSPy is simply a library he uses.
• AI enablement numbers: User’s own lectures achieved 100% application rate. The 1800+ engineers number is about a vendor collaboration the user coordinates, NOT the user’s direct training audience.
• Country names: User said “global Rocket locations” but did NOT list specific countries. Do not fabricate country lists.
• ICDL 2025: Raunit IS 2nd author — strong position.
• Google Scholar: Found and verified — https://scholar.google.com/citations?user=abZZ14QAAAAJ — uncommented in CV LaTeX and added to website homepage.
• RL Regulation project: Ongoing/confidential — descriptions kept general (MuJoCo, multiple policy algorithms, exploration-regulation dynamics).
• Code availability: Most code is NDA’d or under active research grants. User may create 1-2 public repos but can’t share most work.
• PiRat uses ROS2 (NOT ROS1 as might be assumed from older papers).
• Two resume versions: 1-page industry resume focuses on high-impact bullets and keywords; multi-page CV expands everything for academic audiences.
• IBM internship: Removed as a work experience section on industry resume; Aspera test suite is a project there. Full IBM section retained on CV.
• AI Classroom placement: Listed under Research Consultant role (current) on industry resume; could fit under either Staff or Consultant role per user.
• AI enablement numbers: User’s own lectures achieved 100% content application rate. The 1800+ engineers number is about a vendor collaboration program the user is coordinating, NOT the user’s direct training audience.
• Country names: User said “global Rocket locations” — specific countries (US, India, UK, Czech Republic, China) were NOT provided by user and should not be listed.

Links & References

Raunit’s Properties

• Website: https://www.raunitkohli.com/
• GitHub: https://github.com/kohlir2020
• LinkedIn: https://linkedin.com/in/raunit-kohli/
• Email: raunitkohli@gmail.com
• Phone: 774-262-9233

Publications

• ICDL 2025: https://ieeexplore.ieee.org/document/11204421
• Frontiers 2022: https://doi.org/10.3389/fpsyg.2022.897603
• BCI 2025 Abstract: https://openlib.tugraz.at/download.php?id=686289f5ddb80&location=browse

Rocket Software

• EVA Product: https://www.rocketsoftware.com/en-us/products/eva
• SmartChat Brochure: https://www.rocketsoftware.com/sites/default/files/resource_files/smart-chat-brochure.pdf
• NextGen Academy: https://www.rocketsoftware.com/en-us/nextgen-academy
• ACE Paper (ICLR 2026, NOT authored): https://arxiv.org/abs/2510.04618

Research

• Chiba Lab Website: https://www.chiba-lab.org/ (unfinished)
• PiRat IROS 2018 Paper: https://doi.org/10.1109/iros.2018.8594060

Local Directories (for code context)

• NeuroPos Pipeline: /Users/raunit/Desktop/ChibaLab/NeuroPosHb/
• PiRat System: /Users/raunit/Desktop/ChibaLab/pirat/

Job Posting Companies Researched

• Figure AI (robotics ML)
• Skild AI (foundation models for robotics)
• NVIDIA Isaac Lab (robotics research)
• Apple (robotics & automation ML)

Graduate Programs Researched

• CMU Robotics Institute
• Stanford CS (AI track)
• MIT CSAIL
• UC Berkeley BAIR

Known Website Issues (Status as of April 2026)

• ✅ PiRat RL project on research.html — filled with content
• ✅ VEX Robotics on extracurricular.html — expanded with PID/splines/odom/sensor fusion
• ✅ index.html “Please hire me!” text — updated to professional copy
• ✅ Google Scholar — added to homepage social links and CV header
• All major website issues resolved. Website, resume, and CV are in sync.