Software Architect & Engineering Leader (ex-Dolby) | 20+ years building mobile, media, streaming and cloud platforms | Now building agents and evaluation harnesses for the same kind of problem

I'm a technology leader and software architect based in Sydney. My career runs the full media stack, from wavelet codecs on DSPs and FPGAs, through HLS/DASH ad-insertion platforms, to ultra-low-latency WebRTC SDKs on Android. I've moved between hands-on engineering, architecture, and leadership at startups and global technology companies.

• 🔬 Current Focus: Agents and evaluation for problems that need real measurement, not demos — every claim cited, every trade-off on a frontier
• 🚀 Active Projects: perfettoagent · superplayer · pareto-eval
• 📚 Learning: AI Agents course by Ed Donner
• 🏗️ Background: Software architecture, platform strategy, SDK design, and performance engineering (startup time, jank, memory, latency)
• 💼 Experience: Dolby Laboratories (Senior Staff Architect) | AdSparx, acquired by Discovery (VP Engineering) | CCentric, acquired by EY | Co-founder & CEO, Einsteiner Technologies
• 🎓 Education: M.E. & B.E. in Electronics and Telecommunications | Product Management, Stanford
• 🇦🇺 Open to: Senior architecture, engineering leadership, AI-enabled product, and fractional CTO roles in Australia

Languages

• Kotlin, Java, TypeScript, C, C++, Python, Swift

Mobile & Client Platforms

• Android, iOS, React Native, Jetpack Compose, JNI, NDK, AndroidX Media3
• Profiling and tracing with Perfetto and Android Studio Profiler

Media & Streaming

• WebRTC, HLS, MPEG-DASH, Smooth Streaming, RTSP/RTP/RTCP, RTMP, MoQ
• H.264/AVC, JPEG2000, wavelets, FFMPEG, MP4Box, ABR, CTA-2066 QoE
• DRM: Widevine, PlayReady, Common Encryption

Cloud & Backend

• AWS (EC2, S3, CloudFront, RDS, ElastiCache, Elastic Beanstalk), Azure Media Services
• REST, microservices, pub-sub, Node.js

AI & Evaluation

• LLM agents and tool use, LLM-as-a-judge calibration (Cohen's kappa), Anthropic & OpenAI APIs
• Agent evals on planted ground truth, uv, pytest, GitHub Actions

LLM agent for Android performance regressions | Cited diagnosis, verified by code

• Takes two Perfetto traces and a git range, returns which metric regressed, by how much, which commit caused it, and the trace rows that prove it
• Every claim cites a trace query or a commit, and a deterministic verifier re-runs each citation before anything is written, dropping claims whose citation fails
• Covers slow startup, allocation-churn jank, main-thread blocking, layout thrash and memory leaks
• Measured on planted regressions, 20 cases across two apps, each run three times: at xhigh effort, 93% detection, 100% attribution, 96% citation validity, at $0.044 per trace
• Tech: Python, Perfetto trace processor SQL, git tooling, OpenAI & Anthropic APIs
• 📊 Full eval results

A production playback layer on AndroidX Media3 | Policy, resilience, observability, lifecycle

• Not a fork and not a new player: SuperPlayer is a Media3 Player, so existing PlayerView, MediaSession and Compose surfaces take it unchanged
• Content has a stable id instead of a URL, so the cache key, CMCD session, telemetry row and notification all agree on one identity
• 14 modules behind one seam each: ABR, cache, preload, resilience, DRM, offline, TV, diagnostics, realtime, Media over QUIC
• QoE measured to CTA-2066, with every departure from the standard cited
• 18 architecture decision records arguing the design, and a committed benchmark that reports where the adaptive policy lost
• Apache-2.0 · Kotlin · pre-1.0, unpublished by design — the decision records are the point
• Tech: Kotlin, AndroidX Media3, Gradle convention plugins, Robolectric

Multi-axis evaluation for LLM agents | Quality × Cost × Latency

• Reports quality, USD cost and p95 latency together as a Pareto frontier instead of one scalar score
• Borrows the rate-distortion curve from video compression: a configuration is dominated when another beats it on every axis
• Judge-calibration gate: measures LLM-judge vs human agreement with Cohen's kappa before any judge score counts
• CI gate asks "did this change fall behind the frontier we already had?", not just "did quality drop?"
• Tech: Python, uv, Anthropic & OpenAI SDKs, pytest
• 📄 Write-up · Design decisions

• Millicast Android SDK (Dolby): ultra-low-latency WebRTC streaming on Android in Kotlin, Java and JNI
• OptiView Player (Dolby): HLS/DASH streaming with ads and analytics for Android and React Native
• DolbyON / Capture SDK (Dolby): audio/video capture, encode, decode, transcode and editing pipeline for Android
• Dynamic ad-insertion platform (AdSparx, acquired by Discovery): client SDKs, media servers, manifest manipulation, transcoding and linear ad detection
• ABC iView for Android Mobile and TV: Kotlin (consulting)
• Qwik codec family (iFlect): wavelet-based image and video codecs for mobile, DSP and FPGA

• A curated map of free material matching each course in Stanford's AI Graduate Certificate

• 🏅 US Patent 7,522,774: Methods and apparatuses for compressing digital image data (family: EP1730846)
• 🏅 Patent filings on detecting advertisements in streamed media and on dynamic ad insertion in MPEG-DASH and Smooth Streaming with DRM
• 📖 Wavelet Based Scalable Video Codec, Springer/IEEE ICCCT 2010
• 📖 Real time encoding of H.264 on handhelds, Scalable Video Coding and Applications, and more

I'm always happy to talk about:

• 🤖 LLM agents and how to actually measure them
• 📱 Mobile SDK and platform architecture
• 🎬 Streaming, codecs, and media pipelines
• 🧭 Engineering leadership and technical strategy

Reach out via:

• 📧 ramesh130@gmail.com
• 🔗 LinkedIn
• ✍️ Medium

Thanks for visiting! 🚀