Services
What I get hired to build
AI that replaces a process, not just a prompt box. Often with a live voice or video layer, always with the system around it.
AI that does the job
Not a chatbot bolted onto the corner of your product. A system that owns a workflow a person used to run, start to finish.
- Agents that teach, interview, assess, and decide — not just answer questions
- Multi-step flows with memory, state, and progression across sessions
- Evaluation and scoring you could defend to the person being assessed
Live voice & video agents
When the agent has to talk to someone in real time, the hard part is everything that happens when the network, the model, or the user misbehaves.
- LiveKit and WebRTC session lifecycle, reconnection, and low-latency streaming
- Turn-taking and interruption handling that doesn't feel like a phone tree
- Several agents handing off inside a single live session
The system around the agent
An agent is one component. The product is everything holding it up — and that is usually where these projects actually fail.
- FastAPI and Postgres services, background jobs, retries, and migrations
- Reporting, analytics, certification, and the admin surfaces people operate from
- Production triage when an AI feature is already live and misbehaving
Not sure which of these you need? Describe the problem and I'll tell you straight whether I'm the right person for it.
Work
Selected work
A sample of things I've shipped end to end.
Experience
Where I've worked
Mar 2026 – Present
Software Engineer
BerriBot
- Build an AI tutor that does what a human tutor does: teaches live over voice and video, sets and marks assessments, and moves the learner through the syllabus.
- Designed the learning core end to end: syllabus generation, planner agents, coding assessments, mock interviews, verification pipelines, and automated course progression.
- Built the backend on FastAPI, PostgreSQL, and Supabase — APIs, background jobs, retry paths, migrations, and deployments.
- Built AI proctoring with video capture, live monitoring, secure storage, and candidate evaluation.
- Diagnose and fix production failures across voice pipelines, agents, and enterprise deployments — the work that turns a demo into a product teams rely on.
Mar 2025 – Feb 2026
Full Stack Engineer
SkillSync
- Designed and shipped an AI interviewer that replaces the human screening loop — multi-round spoken interviews conducted live over LiveKit and WebRTC.
- Orchestrated several AI agents across screening, technical, and behavioural rounds inside a single live session.
- Owned the real-time voice layer: session lifecycle, low-latency audio streaming, reconnection, and the failure modes that only appear in production.
- Built FastAPI services for session coordination, adaptive question generation, and scoring pipelines.
- Led a database migration off the legacy schema with minimal production downtime, working directly with the founders.
Jul 2024 – Aug 2024
Generative AI Intern
EY · Gurugram
- Built a bench management portal in React to improve utilisation of unassigned staff and reduce hiring spend.
- Built an AI assistant over the internal database using FastAPI, the Gemini API, and fine-tuning.
- Shipped the interactive reporting layer — filterable workforce statistics and charts.
Nov 2023 – May 2024
Full Stack Developer Intern
Ransh Innovations · Remote
- First engineering hire at an early-stage startup, building the product end to end.
- Designed and hardened REST APIs for performance and security, with authentication and data protection.
- Worked directly with the founder on schema design and system architecture.
About
How I work
The interesting AI work right now is not a chatbot bolted onto a product. It is a system that takes over a job a person used to do, end to end. That is what I build: an AI tutor that plans a syllabus, teaches live, sets the assessment, marks it, and moves the learner on. An AI interviewer that runs screening, technical, and behavioural rounds inside one live session and scores the result. It means the live voice and video layer over LiveKit and WebRTC, the agent orchestration and evaluation behind it, and the unglamorous rest — FastAPI and Postgres services, background jobs, retries, migrations, reporting, and the admin surfaces someone actually runs it from. I take this work from an ambiguous first conversation to something that still holds up on its thousandth session, and I stay for the part where it breaks.
New Delhi, India · working with teams remotely
Capabilities
The stack behind the work
AI & Agents
Real-time & Voice
Backend
Data
Frontend
Infrastructure
Contact
Tell me about your project
Describe what you're building and where it's stuck. I'll come back with whether I can help, how I'd approach it, and what it would take.