Skip to content
Mohith Ram Narendra Babu

Software Engineer · UC Berkeley EECS

Mohith Ram Narendra Babu

I build software end to end, from the interface down to the APIs and data behind it. Currently studying EECS at UC Berkeley.

Where I've built

  • AWS
  • Fortinet
  • Maserati
  • CKTH
  • Mustang Math

A project, step by step

Pick a project. Watch what it actually does.

Select a project
Project

Senior care platform

What it does

A caregiver speaks to the platform instead of navigating it. It transcribes what they said, answers through a voice agent, and keeps the health and concierge dashboards in sync for the people watching over them.

  1. 01INPUTA caregiver speaks a request out loud.
  2. 02TRANSCRIBEWhisper turns the audio into text in real time.
  3. 03RESPONDThe voice agent answers, and Fish Audio speaks it back.
  4. 04SYNCSupabase Realtime pushes the update to every dashboard at once.
  5. 05OUTPUTHealth and concierge views reflect the change, scoped by role.

What I built

  • Next.js
  • TypeScript
  • Supabase
  • Whisper
  • Fish Audio

Key decisions

  • Real-time voice interaction with Whisper STT and Fish Audio TTS.
  • Supabase Realtime with role-based access behind both dashboards.
  • Large fonts and simplified navigation, built with Tailwind and shadcn/ui.

Measured result

  • Syncs 10+ wellness metrics across health and concierge dashboards.
WELLNESS METRICS SYNCED
10+

Stream assistant

What it does

Watches where you are looking during a live stream and switches the OBS scene to match, so a solo broadcaster can run a two-camera show without touching the controls.

  1. 01INPUTThe webcam feed streams in frame by frame.
  2. 02TRACKMediaPipe locates the eyes and estimates gaze direction.
  3. 03STABILISEA 10-frame stability check and debounce reject momentary glances.
  4. 04DECIDEA sustained look at the second camera becomes a switch decision.
  5. 05OUTPUTThe OBS WebSocket API changes scene in under 200ms.

What I built

  • Python
  • OpenCV
  • MediaPipe
  • OBS WebSocket

Key decisions

  • Real-time eye tracking built on OpenCV and MediaPipe.
  • OBS scene control driven through the WebSocket API.
  • Debouncing plus 10-frame stability checks to suppress spurious switches.

Measured result

  • Sub-200ms automatic camera switches during live streaming.
  • Over 90% fewer false camera switches under live streaming conditions.
SWITCH LATENCY
<200ms
FEWER FALSE SWITCHES
90%
CAMERAS CONTROLLED
2

Data platform

What it does

Collects sensor readings off the robot during field testing and puts them on a live dashboard, so a failed autonomous run can be read off a chart instead of reconstructed from logs.

  1. 01INPUT10 sensors emit readings throughout an autonomous run.
  2. 02INGESTPython pipelines collect the stream and write it to InfluxDB.
  3. 03PROCESSOver 100K data points are aggregated into route and accuracy series.
  4. 04VISUALISEGrafana renders 50+ metrics as live dashboards.
  5. 05OUTPUTA failed run is diagnosed in under 1.5 hours instead of roughly 5.

What I built

  • Python
  • InfluxDB
  • Grafana

Key decisions

  • Grafana telemetry covering 50+ robot metrics across 10 sensors.
  • Python and InfluxDB pipelines to ingest and process field-test data.
  • Dashboards surfacing route efficiency and navigation accuracy.

Measured result

  • Autonomous debugging time cut from roughly 5 hours to under 1.5 hours.
  • Over 100K data points processed during field testing and validation.
METRICS TRACKED
50+
SENSORS
10
DEBUG TIME
5h → 1.5h
DATA POINTS
100K+

Where I've worked

Where the work actually happened.

  1. 01.

    Software Development Engineering Intern

    Amazon Web ServicesSeattle, WA · June 2026 — August 2026

    Cut agreement lookup time to under 2 seconds by replacing manual search with a Python Bedrock agent mapping queries to 5 APIs. Caught 80% of regressions pre-deployment by replaying 61 labeled queries across 10 workflows, enforced a read-only API allowlist with 22 tests, and reduced lookup tickets 45% across 5 teams.

  2. 02.

    Contract work

    Technical Consultant

    FortinetBerkeley, CA · August 2026 — Present

    Built ranking pipelines scoring 100K+ accounts on funding, hiring, and sector-growth signals, and a monitoring platform on FastAPI, PostgreSQL, and React analytics that cut reporting time 60%.

    Data Consultant

    MaseratiBerkeley, CA · February 2026 — July 2026

    Authored a software-defined ownership strategy for the Folgore EVs, benchmarking 15+ luxury and EV brands and running NLP on 10K+ consumer reviews to cut research time by 40%.

    Software Engineer

    CKTHBerkeley, CA · February 2026 — May 2026

    Built 5 Next.js MVP pages and NestJS admin CRUD APIs supporting 1000+ records, with 20+ unit tests to improve staging release stability.

  3. 03.

    Software Engineer

    Mustang MathSeattle, WA · May 2025 — August 2025

    Led development on a test problem management platform used by Stanford and Berkeley, serving 10K+ students. Built a real-time collaborative testing system with a REST API, SvelteKit, and SQL used by 1.5K+ competitors, and optimized UI/UX with lazy loading to support growth to 75K+ monthly active users.

Selected work

Everything else I've shipped.

  • Senior care platform

    GrandFLOW

    A senior care platform with voice agents and health dashboards.

    View repo

  • Stream assistant

    Cue-Bot

    Eye tracking that switches your camera so you do not have to.

    View repo

  • Data platform

    Robotics Telemetry

    Live dashboards that turned field-test guesswork into a readout.

    Private repo

  • Market data

    Kalshi Mispricing Engine

    Reads live prediction-market prices and flags the ones that look wrong.

    View repo

  • Language models

    GPT-2 Inference Engine

    Runs GPT-2 on a laptop CPU with no ML framework underneath.

    Private repo

Get in touch

If you are working on something interesting, I'd like to hear about it.

mohithn@berkeley.edu