INDEX / 2013 — 202637.7749°N 122.4194°W

Rajasekhar Josyula

Photons in, decisions out.

00 // IDENT

Role
Staff Software Engineer, Tesla AI
Location
San Francisco, CA
Focus
Real-time perception · NN training · C++20 · CUDA

01 // ABOUT

I am a Staff Software Engineer at Tesla AI, building sub-10ms photon-to-NN camera pipelines for production cars and Optimus robots, and training the neural nets used for driver monitoring and cabin monitoring. I lead sensor bring-up and real-time calibration, and have been hand-crafting low-level C++ for a decade to keep perception airtight from prototype labs to Robotaxi production.

02 // SPECTRUM

SPECTRUM // ACCUMULATED_CAPABILITY_ANALYSIS
T_OBS: 2026.0INTEGRATOR: ACTIVE
E(t) = Σᵢ aᵢ·Πᵢ(t) · dE/dt ≥ 08/8 LIVEE25%50%75%dE/dt — the rate of learningΠᵢ(t) = 3u² − 2u³ · u = clamp((t − tᵢ)/λ, 0, 1)1992200020102020nowaᵢ(τ) = aᵢ·Πᵢ(τ)τ = nowω₀CURIOSITY'92ω₁CIRCUITS'09ω₂ROBOTICS'13ω₃DATA'16ω₄PERCEPTION'17ω₅REAL-TIME'20ω₆DEPLOYMENT'21ω₇TRAINING'22S(ω,t) = |X(ω,t)|² · onsets never decayω₀ω₁ω₂ω₃ω₄ω₅ω₆ω₇'92'00'10'20nowΠᵢ(t) is monotone — nothing learned is ever lost
Fig. 1 — Capability as accumulated energy. Each component switches on with a smoothstep onset Πᵢ(t) and never decays, so E(t) only rises. Sweep the curve to scrub τ and read each component's amplitude at that instant; hover a bin to isolate its contribution, click to mute.

03 // PROJECTS

  1. xSLAM: Real-Time SLAM in the Browser_

    2025

    Real-time SLAM visualization running 100% in the browser. Autonomous drone simulation, 3D Gaussian Splatting, Monte Carlo Localization, and Graph SLAM visualization with no backend required.

    xSLAM // SPATIAL_ESTIMATOR_ACTIVE
    FPS: 60.0T_PROC: 4.8MS
    An interactive Bayes filter: hover graph nodes to inspect local covariance ellipses, trigger 2D LiDAR raycasts, and highlight Information matrix components. Click nodes to inject factor constraints (x4 loop closure).
    SourceLive DemoRead Case Study

  2. copypaste.fyi: A Secure Clipboard Relay_

    2024

    Clipboard relay for distributed teams: drop text or secrets on one machine and retrieve them securely anywhere via short-lived, single-use tokens. Built for tmux-first workflows with deterministic latency on the edge.

    COPYPASTE // SECURE_TUNNEL_ESTABLISHED
    E2EE: X25519AEAD: CHACHA20
    The protocol, step by step — hover entities to inspect variables (A public keys, Relay storage structure, B decay triggers). Click B to force self-destruction sparks, click the TTL bar to reset the countdown.
    SourceLiveRead Case Study

04 // EXPERIENCE

  1. Staff Autopilot Software Engineer_

    2022.05 — present

    Trains multi-task vision models for driver monitoring and cabin monitoring — drowsiness detection, gaze and attentiveness estimation, EU homologation (DDAW/ADDW) — and designs sub-10ms photon-to-NN camera pipelines for FSD vehicles, Optimus, Cybercab, and Robotaxi.

  2. Sr. Software Engineer, Autopilot_

    2020.02 — 2022.05

    Led camera bring-up and Sony IMX sensor validation for Model 3/Y/S/X. Built vehicle-level image-quality pipelines, maintained the production C++ capture stack, and shipped closed-loop IQ tuning via OTA updates across the global fleet.

  3. Sensor Software Engineer, Autopilot_

    2017.05 — 2020.02

    Led factory camera calibration for vision-only Autopilot using ChArUco diamond boards. Optimized embedded ISP pipelines and owned camera intrinsic generation at contract manufacturers.

  4. Software Engineer, Quality_

    2016.01 — 2017.05

    Product improvement for Model X with Python, Tableau, and Flask tooling. Root-cause analysis on authentication failures and data pipelines correlating fleet telemetry with external quality metrics.

05 // EDUCATION

  1. M.S. Information Systems & Security_

    2017

    University of the Cumberlands. GPA 4.0 · Cybersecurity focus.

  2. M.S. Electrical & Electronics Engineering_

    2016

    Lamar University. GPA 3.6 · Embedded systems focus.

  3. B.E. Electrical Engineering_

    2013

    Amrita University. Robotics focus.

06 // UPLINK

Email
qxlsz@duck.com
GitHub
github.com/qxlsz
LinkedIn
in/rajasekharjosyula
X (DM)
@qxlsz
Signal
qxlsz.6174
Matrix
@qxlsz:matrix.org

07 // TRUST_ANCHORS

PGP · RSA · primary

AC7CEA1EA4E8AEB0450954B7715483B8A3E539CB

SSH · ed25519

AAAAC3NzaC1lZDI1NTE5AAAAIE9YHF9s3Z7Yp4czFZc9ox96D65V6pZeRAEH6kT6Gatg

verify: keybase.io/qxlsz · github.com/qxlsz.keys