Voice AI Agents
Production agents over telephony & WebRTC: streaming STT → LLM → TTS, tool-calling, hot-swappable personas.
Read howFounding Engineer @ Vaani AI
Voice AI agents,blazing-fast systems &full-stack products- all in one engineer
I'm Mohammad Palla. I build production voice AI that talks back in under 300ms, make slow code 4× faster, and ship everything in between: schema to SDK.
Hi! I ship voice agents that pick up the phone.
Player 1▶ Press start
Mohammad Palla
Class: Founding Engineer · XP 3+ yrs
4.45× faster indexing−58% peak RAM−67% CPU per indexsub-300ms voice RAG50% cheaper inference90,000 fuzz probes · 0 panics57,000 files byte-identicalAPIs up to 80% faster
What I build
Six skill trees, all maxed out in production
Performance & Systems
Profiling-led speedups and native Rust (PyO3 + tree-sitter) kernels. 4.45× faster on 58% less memory.
See proofFull-Stack Product
React, Next.js, FastAPI, Django & Node: from database schema to pixel-perfect UI, owned end-to-end.
See workAPIs, SDKs & Docs
Public APIs, SDKs and Mintlify docs that turn an internal platform into something other developers love.
See workRAG & GenAI
Retrieval pipelines tuned for real-time conversation: 50% cheaper inference at 90% retrieval accuracy.
Read howInfra & DevOps
Blue-green deploys, Docker, Nginx, Celery + Redis, Datadog, and zero-drama AWS → Azure migrations.
See workAbout Mohammad
I'm a founding engineer who's happiest where hard problems meet real users. My title has said “Founding Engineer” for a while, but the work has been all over the place, in the best way possible.
I've architected voice AI systems from telephony to agent orchestration, designed the schemas and concurrency systems underneath them, and gone from consuming APIs and SDKs to designing and shipping our own. On contract, I took a code indexer from 10 minutes to 2, with byte-identical output.
When I'm not in the editor, I'm on calls with customers, recording walkthroughs, hosting hackathons, or, apparently, acting in launch videos.
Hyderabadi ♥
How I work
The loop that turns slow & flaky into fast & proven
Profile
Measure before touching code. py-spy, flamegraphs and Datadog over guesswork.
Build
Own it end-to-end: schema, API, UI, infra, docs. Ship it, then make it fast.
Verify
Differential tests, fuzzing and parity harnesses. Byte-identical or it didn’t ship.
Quest log
The main story: full-time roles & internships, newest first. Scroll to walk the path.
Vaani AI Research
Oct 2025 - PresentFounding Engineer
Full-time · Bangalore
Built the voice AI platform from the ground up: telephony, WebRTC, agents, APIs, SDK and docs.
Loot
- Voice AI infrastructure: production-grade, end-to-end voice agents over telephony and WebRTC; benchmarked providers, implemented streaming, built the full tool-calling pipeline.
- RAG for voice: re-engineered retrieval to cut inference cost 50% and end-to-end latency to sub-300ms at 90% retrieval accuracy.
- Multi-agent transfers: in-call hot-swaps of persona, prompt and tools within one session, with per-session caps to prevent runaway loops.
- Developer platform: designed and shipped the public API & SDK, authored the Mintlify docs, owned the React frontend and FastAPI backend.
- Core systems: callbacks & campaigns, APIs optimised by up to 80%, nearly the entire DB schema, concurrency & agent management, auth and middleware.
- Async infrastructure: post-processing pipeline from scratch on Celery, Celery Beat and Redis.
- Beyond code: forward-deployed with customers, sales support and product input that lifted engagement.
PythonFastAPIReactWebRTCCeleryRedisPostgreSQLFindaTutor.ch
Mar 2025 - Oct 2025Full-Stack Engineer
Full-time · Remote (Switzerland)
Led the rebuild of a Swiss tutoring marketplace, including a live MySQL → PostgreSQL migration.
Loot
- Legacy migration: led the full codebase and database migration from MySQL to PostgreSQL with data integrity throughout.
- New product version: owned the Node.js + React rebuild end-to-end, from architecture to production.
- Booking & calendars: scheduling, availability and conflict resolution for tutors and students.
Node.jsReactPostgreSQLMySQLSigiq
Sep 2024 - Mar 2025Full-Stack Engineer
Full-time · Gurugram
Built an EdTech voice tutor end-to-end and hardened the backend & infra underneath it.
Loot
- AI voice agent: Deepgram (STT) + Anthropic tool-calling + ElevenLabs & Cartesia (TTS) in a low-latency production pipeline, plus the streaming-audio frontend.
- Backend: Django, PostgreSQL, Redis, Celery; resolved migration conflicts between two services sharing a schema.
- DevOps: blue-green deployments from scratch, Docker, NGINX, Datadog, and an AWS → Azure migration.
- Frontend: legacy app → Tailwind PWA; new greenfield frontend architecture.
DjangoDeepgramAnthropicElevenLabsAzureOne Immersive
Apr 2024 - Jul 2024SDE Intern
IIIT Hyderabad
Full-stack owner of ArenaExplore.com, from Next.js frontend to production deploys.
Loot
- ArenaExplore.com: NextJS (Zustand, Tailwind) frontend and Node/Express + MongoDB/Firebase backend.
- Cross-platform auth: integrated Unity authentication for the mobile app; deployed on AWS with Nginx and pm2.
Next.jsExpressMongoDBUnitymaths.ai
Jun 2023 - Apr 2024SDE Intern
Remote
Shipped a React Native MVP in one week with zero prior mobile experience, plus all the GenAI features.
Loot
- Mobile, web & backend: React Native app (MVP in 1 week), NextJS web app, shared Node/Express backend.
- Generative AI: low-cost, low-latency RAG with Llama-Index and Groq.
React NativeNext.jsLlama-IndexGroqTraining grounds
2020 - 2024Chaitanya Bharathi Institute of Technology
B.Tech, Computer Science Engineering · GPA 8.40 · Hyderabad
Operating systems, data structures, algorithms, AI/ML, networking, databases, full-stack & mobile development.
Bounty board
Contract gigs: short, sharp engagements with measurable loot
★ Bounty · RemoteCleared
GraphifyYC S26
Performance & Systems Engineer Aug 2026
Reward claimed: Made a code-graph indexer 4.45× faster on 58% less memory, erasing a 3.9–7.9× gap to the leading commercial indexer.
- 4.45× faster indexing: cold index of the Bun monorepo (25,725 files) from 10m27s → 2m21s (−77.5%), turning a CI-blocking job into interactive latency and unblocking per-commit indexing.
- 58% less RAM, 67% less CPU: peak memory 5.53 GiB → 2.30 GiB while lifting parallel utilisation from 129% to 188%.
- Erased a 3.9–7.9× competitive gap: now at wall-clock parity with the leading commercial indexer on 42% less memory and 15% less CPU.
- Native Rust extraction engine across 10 languages (PyO3 + tree-sitter), fail-open per file, proven byte-identical across ~57,000 files in ~30 OSS repos.
- Profiling-led, not guesswork: deleted 5.2M redundant Path constructions and 5.4M lstat syscalls; one loop-invariant set() rebuild was 71% of a pass (50.6s → 4.7s).
- Verification engineering: differential-parity & fuzz harnesses (~90,000 probes, zero panics); grew the suite 5,302 → 5,530 tests and surfaced a latent cache-invalidation bug.
RustPyO3tree-sitterPythonpy-spy
★ Bounty · RemoteCleared
Confidential clientYC F24
Voice AI Engineer Mar 2025
Reward claimed: One-month engagement to make a production Vapi voice agent faster and more reliable.
- Agent redesign: audited and reworked agent design, system prompts and tool-calling for reliability.
- Latency & quality: tuned STT / LLM / TTS selection and streaming; improved turn-taking and interruptions to cut dropped and misrouted calls.
VapiPromptingTool-calling
Side quests
Achievements unlocked outside the code editor
★ Unlocked
Movie Star (Sort Of)
Acted in Vaani's launch video. Not something I had on my engineering-career bingo card.
★ Unlocked
Host Mode
Organized and hosted a hackathon, including anchoring the entire event.
★ Unlocked
Security Sensei
Hosted an online cybersecurity workshop for external participants.
★ Unlocked
Field Agent
Visited customer offices and helped them integrate Vaani, forward-deployed style.
★ Unlocked
Hype Engineer
Spoke with influencers and worked on PR for the launch.
★ Unlocked
Loom Legend
Recorded countless Loom walkthroughs to explain features and document the platform.
★ Unlocked
Code Guardian
Reviewed code, wrote documentation and kept the platform walkthroughs current.
★ Unlocked
Speedrunner
Learned React Native and shipped a mobile MVP in a single week.
Inventory
Side projects, packages & experiments I've shipped
Prove It
A GitHub PR reviewer that executes code instead of just reading the diff.
- Clones the PR into an isolated Solari sandbox
- Installs deps & runs the existing test suite
- Generates repro scripts with Gemini Flash
- Verdict grounded in what actually ran
React/Vite · Vercel Functions · GitHub API · Gemini
Try it live ↗pyapimon
Drop-in API monitoring middleware for FastAPI, Django and Flask.
- Captures request/response data locally
- Rich CLI: stats, dashboards, ASCII graphs
- LLM-powered insights on your traffic
- Zero external services required
Python · FastAPI · Django · Flask · Rich
View on PyPI ↗Hyderabadi Lang
A programming language that speaks Hyderabadi, published on npm.
- Forked from open-source bhailang
- Custom keywords & grammar
- Interpreter in TypeScript
- Bilkul mast language, miyan
TypeScript · npm
View source ↗RAG without OpenAI
Low-cost, low-latency RAG microservice on Groq + Llama-Index, no OpenAI keys.
- Dug into Llama-Index source to unlock custom prompts + memory
- Runs on an 8 GB laptop
- Hugging Face embeddings
Python · Flask · Groq · Llama-Index
Read the postPetrol Pump CRM
Custom CRM for a petrol pump with a fault-tolerant design built from scratch.
- System & schema designed from zero
- Robust, fault-tolerant architecture
Next.js · Node.js · MongoDB · Express
MindTherapy
Therapy booking app: patients book, doctors manage schedules.
- Separate auth flows for patients & doctors
- Appointment booking & schedule views
Next.js · Node.js · MongoDB · Express
Skill tree
Equipped items. Hover to inspect.
Languages
- Python
- JavaScript
- TypeScript
- Rust
- C/C++
- SQL
- Bash
Frameworks
- React
- Next.js
- React Native
- FastAPI
- Django
- Flask
- Express
- Node.js
Voice & AI
- WebRTC
- LiveKit
- Vapi
- Pipecat
- Deepgram
- ElevenLabs
- Cartesia
- AssemblyAI
- Anthropic
- Gemini
- Groq
- Llama-Index
Data & Infra
- PostgreSQL
- MySQL
- MongoDB
- Redis
- Celery
- Docker
- Nginx
- Datadog
- Mintlify
Platforms
- Linux
- AWS
- GCP
- Azure
- Vercel
Soft skills
- Leadership
- System Design
- Product Thinking
- Public Speaking
- Time Management
Dev log
Stories from the trenches of voice AI, performance & the web
Are you open to freelance or contract work?
Yes, selectively. I take short, high-leverage engagements: performance audits, voice AI agents, and API/SDK/platform work. Email me what you are building and where it hurts.
What does a "founding engineer" actually do?
Everything that needs doing. For me that has meant architecting voice AI from telephony to agent orchestration, designing the database and concurrency systems, shipping a public API + SDK + docs, and also talking to customers, hosting a hackathon and acting in our launch video.
Can you make my voice agent faster?
Probably. Most latency hides in provider choice, missing streaming, turn-taking and retrieval. I have taken voice RAG to sub-300ms end-to-end and tuned production Vapi agents to cut dropped and misrouted calls.
How do you approach performance work?
Profile first, then fix the biggest thing, then prove nothing changed. On Graphify that meant py-spy attribution, a Rust (PyO3 + tree-sitter) kernel, and differential + fuzz harnesses showing byte-identical output across ~57,000 files.
What is your go-to stack?
Python (FastAPI/Django) or Node on the backend, React/Next.js on the frontend, PostgreSQL/MongoDB + Redis + Celery for data and jobs, and Rust when the hot path needs it.
Where are you based?
India: Hyderabadi at heart, working with teams in Bangalore, Switzerland and the US. Remote-friendly and comfortable across time zones.