●Open to full-time roles · Available for freelance

Akhil Lavudya

AI / ML Engineer — IIT Guwahati

Final-year B.Tech at IIT Guwahati. I build agentic AI systems, ML models, and the product around them — then run them against real users and real money. Five automation systems in production or daily use; 313 cold calls logged on my own lead engine.

See the systems →Resume GitHub LinkedIn

5
systems shipped: 313
cold calls logged: 2
first-author papers in prep: 153K
lines across those systems

IIT Guwahati, India

01Systems

Systems I ship

All 5 systems →

Five AI automation products, built end to end. Each one has a database, a long-lived worker, a tenancy model, and a specific failure mode it was designed around. The numbers below are counted out of the repositories, not estimated.

Nova Reception

AI phone receptionist · multi-tenant SaaS

In build

A phone number that answers, understands, and writes to a database.

An AI receptionist for Indian service businesses. It takes the call, answers from the business's own knowledge base, books against a real calendar, captures the lead, and hands off to a human when it should — for one salon or for forty under a single login.

TypeScript LOC: 48K
Agent tools: 17
Test cases: 543
Workspace packages: 9

TypeScript (strict)
Next.js 15
Fastify + ws
Prisma
PostgreSQL 16
pgvector

Read the case study →

OutreachEngine

Cold-email engine · find → write → send → book

Shipped

Optimises one number: positive replies per week. Everything else is diagnostic.

A single-operator outreach system. It finds and stores prospects, researches them, writes a personalised first touch, sends from a real Gmail mailbox on a warm-up schedule, detects and classifies replies, books the meeting, and reports the conversion.

TypeScript LOC: 53.5K
Test cases: 1,017
Migrations: 17
Phases shipped: 11 / 11

Next.js 16
TypeScript (strict)
Drizzle ORM
PostgreSQL 16
Zod
Vitest

Read the case study →

Lead Generator + Outreach Cockpit

Local-business lead engine · call & WhatsApp cockpit

In daily use

The system I actually run my calling day on.

Two programs over one database. The generator finds local businesses in a city, grades how weak or absent their web presence is, and writes an Excel call sheet you can update in three seconds while the phone is still at your ear. The cockpit reads the same database, drafts Hinglish WhatsApp follow-ups with Gemini, and walks a capped daily send list — then tracks replies, callbacks and outcomes.

Leads generated: 1,760
Cities covered: 9
Cold calls logged: 313
Businesses rung: 216

Python
Playwright
FastAPI
SQLite
React 19
Vite

Read the case study →

Instagram Outreach

Instagram prospecting · search, triage, draft

Shipped

Takes it to the last safe step, then stops.

Finds business accounts by searching for them in a real browser, records what each account says about itself, probes the link in its bio, triages the results on a board, and drafts a personalised DM for the ones worth messaging.

TypeScript LOC: 19.6K
Worker jobs / crons: 3 / 2
Pipeline stages: 5
Sends performed: 0

Read the case study →

shorts-clipper

Video pipeline · long-form in, vertical Shorts out

Working end to end

Watchlist-driven, rights-gated, resumable.

A seven-stage pipeline that discovers candidate videos, ingests one, transcribes it, scores moments against a rubric, renders a captioned vertical clip with a ducked music bed, writes the upload metadata — and then re-fits its own scoring weights against how the clip actually performed.

Python LOC: 10.9K
Tests: 455
Pipeline stages: 7
Network calls in tests: 0

Read the case study →

02Commercial

I sell the work too, not just build it

Most of these systems exist because I needed them. I cold call local businesses in North-East India, and sell and deliver websites off the back of it. Every number here is counted out of the database my own cockpit writes to.

1,760
Leads generated: 313
Cold calls dialled: 216
Businesses reached: 31
WhatsApp opt-ins

How it actually runs

Websites sold and delivered to paying clients, sourced from my own calling — not from a marketplace or an agency's pipeline.
A 40-dial daily target, tracked against actual dials, with callbacks surfaced as a dated queue rather than remembered.
One permanent do-not-call list across every city, honoured in every future run.
The build order follows the phone. Nova Reception exists because the most common thing a salon owner said to me on a call was that they miss calls while they are with a customer.

How the cockpit is built →

The Outreach Cockpit dashboard showing 1,760 leads, 1,566 worth pitching, 216 rung and 109 crossed out, above a dials-per-day chart totalling 313 dials and a panel of callbacks due. — Outreach Cockpit — the dashboard I open before the first call of the day.

03Research

Research at CMC Lab, IIT Guwahati

Full experience →

Two first-author manuscripts in preparation under Dr. Kalishankar Bhattacharyya. Both are under embargo, so these pages carry architecture, method, and validation results — no code and no live demo until publication.

May 2026 - Jul 2026

Materia - Talk to a Materials Simulation Engine

Research Intern · CMC Lab, IIT Guwahati

Embargoed

Built an agentic materials-simulation platform that turns natural-language requests into real scientific workflows: VASP/DFT input generation, structure utilities, ML-potential relaxation, MD, elastic, phonon, and NEB flows across 23 tools.

Built a full-stack AI simulation platform (~26K LOC) with FastAPI, React 19, Electron, Celery/Redis, JWT auth, Fernet-encrypted API-key storage, Docker, and GitHub Actions CI.
Designed a 23-tool native function-calling agent on a Gemini 2.5 Flash -> Groq Llama-3.3-70B -> local Ollama failover chain, validated at 100% tool-selection and 100% argument accuracy (2.09s mean latency) over a curated 39-prompt suite.

AI Engineering
Full-Stack Systems
ML for Science
Research Software

May 2026 - Jul 2026

DKL-BO - Smarter Materials Discovery with Deep Kernel Bayesian Optimization

Research Intern · CMC Lab, IIT Guwahati

Embargoed

Designed and evaluated a CGCNN plus Gaussian-Process active-learning system for pool-based 2D-materials discovery, locating the best candidate in a 3,351-material search space while evaluating only ~3% of the pool.

Designed a Deep Kernel Learning surrogate with a CGCNN graph encoder, Gaussian Process uncertainty, and UCB / Expected-Improvement acquisition.
Located the global optimum within ~3% of evaluations - 6.3x more efficient than random search across the fixed candidate pool.

Data Science
Bayesian Optimization
Graph ML
Research Engineering

04Projects

Engineering & ML projects

All 10 projects →

Full-stack engineering, competition machine learning, agentic GenAI, and analytics — everything here has a demo, a repository, or both. The embargoed research sits in the section above.

Software Engineering

Verdex - Full-Stack Online Judge

Complete & deployed

A LeetCode-style online judge where users solve C++ and Python problems in the browser, submit to a compile-once/run-all-cases judge, and receive AC, WA, TLE, RE, or CE verdicts with leaderboard tracking.

Full-stack judge with FastAPI, React/Vite, SQLite, Docker, GitHub Actions CI, and AWS EC2/ECR deployment.
Runs untrusted submissions in Docker sandboxes with networking off, 256 MB memory, 1 CPU, 64 PID caps, read-only FS, and rate limits.
Ships admin problem authoring, profiles, a leaderboard, Gemini AI code review, pagination, and 18 backend tests.

FastAPI
React
Vite
SQLite
Docker
AWS
GitHub Actions
Gemini

Case study →Watch demo Code

Data Science Competition

NFL Draft Prediction - Gradient-Boosting Ensemble

Deployed - Top 10% (GCI World 2026)

A LightGBM + CatBoost ensemble predicting whether a college athlete gets drafted into the NFL from Combine data, served as a live serverless API with SHAP explanations. Placed in the top 10% worldwide at GCI World 2026 (University of Tokyo, Matsuo Lab).

Reached 0.829 OOF ROC AUC with Optuna tuning (50 trials), 5-fold CV, 3-seed averaging, and an 85/15 LGBM-CatBoost blend.
Used leak-safe fold-internal target encoding for 236 schools and treated missing Combine drills as predictive signal.
Deployed as a serverless Modal /predict API with a Gradio UI, SHAP per-prediction factors, and Weights & Biases tracking.

LightGBM
CatBoost
Optuna
SHAP
Modal
Gradio
W&B
ROC AUC

Case study →Live API Code

GenAI / Agents

Sage - Agentic Multi-Tool RAG Chatbot

Complete & deployed

A deployed LangGraph assistant that decides per turn whether to answer directly or reach for a tool - web search, live stock quote, calculator, or semantic retrieval over your uploaded PDF - and remembers every conversation across restarts.

Built as a thin Streamlit client over a serverless Modal FastAPI backend (5 bearer-authed endpoints, scales to zero).
LangGraph conditional tool-execution graph routes across 4 tools with token streaming; Groq Llama-3.1-8B for inference.
Per-document RAG in Qdrant (thread-scoped payload filter), cross-restart memory via a Neon Postgres checkpointer, LangSmith tracing.

LangGraph
Groq
Qdrant
Neon
Modal
Streamlit
LangSmith
FastAPI

Case study →Live demo Code

05Stack

What I work in

Grouped by what I actually reach for, not by what looks broadest. Everything listed here appears in a shipped system, a research pipeline, or a deployed project on this site.

Languages

Python
TypeScript
SQL
C/C++
JavaScript

Machine Learning & Deep Learning

PyTorch
TensorFlow
Keras
scikit-learn
LightGBM
CatBoost
XGBoost
Optuna
SHAP
OpenCV

GenAI / LLM Engineering

LangChain
LangGraph
RAG
Native function calling
Tool-use agents
Agent eval harnesses
Qdrant
pgvector
LangSmith
Gemini
Groq
Ollama

Data & Analytics

Pandas
NumPy
Matplotlib
Seaborn
Power BI
Weights & Biases

Web & Full-Stack

Next.js (App Router)
React 19
Node.js
Fastify
Express.js
FastAPI
Tailwind CSS
shadcn/ui
Drizzle ORM
Prisma
Zod

Infra, Cloud & Databases

Docker
AWS (EC2/ECR)
Modal
PostgreSQL 16
SQLite
Neon
Redis
Celery
BullMQ / pg-boss
Turborepo / pnpm
Git / GitHub Actions

Testing & Reliability

Vitest
pytest
Playwright (e2e)
Load testing
Multi-tenant isolation tests
Idempotency & replay safety

Automation & Go-To-Market

Playwright browser automation
Gmail API
Google Calendar API
Cal.com webhooks
Telephony (Exotel)
FFmpeg
Cold calling
Client delivery

06Contact

Let's talk.

I'm looking for full-time AI / ML engineering roles, and I take on automation and website work alongside it. Email is the fastest way to reach me — I read everything.

akhillavudya4567@gmail.com