AI voice agents & receptionists
Phone agents that answer, schedule, take intake and route calls, around the clock.
Best for: clinics, law firms and service businesses buried in calls.
Incoming callOrthopedic clinic · after hours
VAD · listening
Illustration
About
I've spent the last two years putting AI in front of real people: patients calling a clinic, clients calling a law firm, customers proving who they are on camera.
I started on backend APIs and CI/CD, brought that discipline into Video KYC and RAG at Code Himalaya, then built a voice agent answering a US hospital's phones at MeetGabbi. There, a wrong answer isn't a bug report. It's a patient left waiting.
I'm finishing my BSc IT (Hons) in Kathmandu, where my final-year project teaches the city's own traffic signals to adapt. I speak Nepali, English and Hindi.
Why me
Why teams hire me
Most AI never makes it past the demo. That last mile is where I work.
Experience
-
Feb — Jun 2026
AI Engineer · MeetGabbi
Cincinnati, USA · remote
- Built a HIPAA-compliant AI voice receptionist for a US orthopedic hospital on an agentic architecture: scheduling, intake, triage and follow-ups.
- Engineered the core voice engine on the GPT Realtime API, with VAD gating against false triggers and multi-speaker turn-taking.
- Delivered production voice AI for ARC Psychiatry and US law firms, and mentored the onboarding team.
-
Nov 2024 — Feb 2026
AI Developer / ML Engineer · Code Himalaya
Kupondole, Lalitpur
- Designed and built Video KYC, intelligent document parsing and conversational chatbots with computer vision, OCR, NLP and LLMs.
- Built AI agents and RAG pipelines with LangChain, LangGraph, ChromaDB and Qdrant, with tool calling across enterprise APIs.
-
Mar — Nov 2024
Junior Backend Engineer · Digital Horizons
Kathmandu
- Built RESTful FastAPI microservices and optimised MongoDB databases on AWS.
- Automated CI/CD pipelines, cutting manual deployment time by 80%.
-
Sep 2023 — Now
BSc IT (Hons) · TechSpire College
Kathmandu
Final year project on AI traffic-signal control. Research Analyst in the college club.
Projects
Kathmandu Smart Traffic · live dashboard● LIVE
Featured · Final Year Projectreal screenshots
Smart Traffic Signal Control for Kathmandu ↗
A temporal graph neural network and MaskablePPO agent controlling 7 junctions on the Tinkune → Singha Durbar corridor in SUMO, inside a safety envelope it can't override.
−48.1%waiting time
+8.4%trips completed
−14.8%CO₂
-
HIPAA-compliant AI Voice Receptionist↗
Production voice agent for a US orthopedic hospital: GPT Realtime engine, VAD gating, multi-speaker turn-taking and Twilio PSTN streaming.
Live for 3+ US clients -
Video KYC↗
Biometric identity verification with ArcFace + RetinaFace matching, Silent-Face anti-spoofing and ID OCR, served from AWS.
97.8% liveness TPR · 90%+ match -
AI Code Assistant↗
GitHub agent with RAG over whole codebases; tool-calling workflows for repo search, summaries, docs and PR analysis.
-
PersonaPlex Full-Duplex AI↗
Deployed NVIDIA's 7B speech-to-speech model across GPU clouds, working through CUDA memory limits and CPU offloading.
16.7 GB model · 4 GPU clouds -
Real-Time Voice AI Agent↗
Kyutai STT/TTS with an LLM dialogue engine: multi-turn context, interruption handling and silence detection.
-
AI-Powered Resume Parser↗
PDF, DOCX and scanned CVs to structured JSON with Gemini Vision OCR and LLM extraction.
95%+ accuracy -
Article Recommendation System↗
Sentence-Transformer embeddings in ChromaDB, async MongoDB and nightly updates, with a Streamlit front end.
-
Suspicious Transaction Reporting↗
Real-time anomaly detection for a fintech app using unsupervised clustering and statistical models.
Testimonials
What people say
Trusted by the people I've shipped with.
Services
For teams and businesses
AI engineering and data science, end to end.
From a question in your data to an AI system in production: I analyse, model, build and ship, then hand you something you can measure. Pick what you need, or all of it.
Agentic workflow automation
LangGraph agents that plan, call your APIs and hand off to humans when they should.
Best for: repetitive intake, triage, follow-ups and back-office work.
RAG knowledge assistants
Chat over your documents, grounded in retrieved sources and tuned to reduce hallucinations.
Best for: support teams, internal wikis and policy-heavy work.
LLM apps, prompts & fine-tuning
Integrate GPT, Gemini, Groq or open models into your product, with prompts and fine-tuning tuned for accuracy and cost.
Best for: teams adding AI features to an existing product.
Computer vision & document AI
Identity verification, liveness checks, OCR and parsing that turns documents into clean data.
Best for: fintech onboarding, KYC and document-heavy operations.
Speech & NLP pipelines
Transcription, summarisation, named-entity recognition and semantic search over text and audio.
Best for: call analytics, meeting notes and search over unstructured text.
Predictive analytics & ML models
Models that predict outcomes from your data, built with scikit-learn, PyTorch or TensorFlow and validated properly.
Best for: teams with historical data and a decision to improve.
Recommendation systems
Content and similarity-based recommenders using embeddings and vector search, refreshed on a schedule.
Best for: media, e-commerce and content platforms.
Anomaly & fraud detection
Unsupervised models that flag unusual transactions or behaviour in real time.
Best for: fintech, payments and risk teams.
Data analysis & dashboards
Clean, explore and explain your data, then ship interactive dashboards that decision-makers actually use.
Best for: founders and managers who need answers from their data.
Reinforcement learning & optimisation
Agents that learn control policies in simulation, with safety guardrails and honest baselines.
Best for: scheduling, control and resource-allocation problems.
Model evaluation & benchmarking
Independent evaluation of an AI system against strong baselines, with retrieval quality, hallucinations and real metrics.
Best for: teams unsure whether their model is actually working.
Model deployment & APIs
Turn a notebook model into a fast, documented FastAPI service on Docker and AWS.
Best for: teams with a working model and no way to ship it.
CI/CD & MLOps
Automated pipelines for testing, versioning, deploying and monitoring models.
Best for: teams losing hours to manual releases.
GPU inference & large models
Run large open models on cloud GPUs by managing memory, offloading and streaming inference.
Best for: teams self-hosting speech or language models.
Ways to work together:Full-time roleContractFixed-scope project
Skills
Agents & LLMs
Voice AI
ML & Vision
Engineering
FAQ
Before you reach out
Questions recruiters and clients ask me.
What roles are you looking for?
AI/ML engineering roles focused on voice AI, AI agents and RAG, where the work ships to real users. I'm also open to contract and fixed-scope projects.
Have you shipped to production, or only built prototypes?
Production. My voice agents ran for a US orthopedic hospital, ARC Psychiatry and US law firms. At Code Himalaya I deployed AI services with Docker, AWS and CI/CD.
Can you work remotely with a US-based team?
Yes. At MeetGabbi I worked remotely from Kathmandu with a team in Cincinnati, collaborating with product and engineering on healthcare and legal features.
Can you handle sensitive or regulated data?
I've built HIPAA-compliant voice systems with compliance-aware conversation design. My Video KYC system kept face embeddings, audit logs and ID documents securely in Amazon S3.
How do you know an AI system is actually working?
I measure it against honest baselines. In my traffic project, every trained model had to beat a random-control baseline, not just fixed timers. When it didn't, I said so and worked out why.
How quickly can you start? Do you have references?
Email me your timeline and I'll reply with my availability. References from past managers and colleagues are available on request.
Ask my AI
Pukar.ai
Interview my AI twin. It has read my whole CV: every role, project, number and skill. Ask what you'd ask me.
Contact
Let's talk
Let's build AI your users actually trust.
Hiring for voice AI, agents or RAG, or have a project in mind? Send me the problem and your timeline, and I'll reply with how I'd approach it.
Or copy my email directly:
Click to copy · or call +977 984-5058740