AI Systems Engineer

Hi, I'm Abheesht

The model hallucinates. The infra breaks. The bill explodes. I am the engineer who prevents all three.

I'm into →

CURRENTLY OBSESSING OVER

01.RoPE

02.KV cache optimization

03.GQA

04.inference engines

TECH ARSENAL

LANGUAGES

PythonC++GoSQL

AI / ML / DL

PyTorchTensorFlowHugging Face TransformersPEFTLoRABitsAndBytesScikit-learnNumPyPandasMatplotlibOpenCV

FRAMEWORKS & ORCHESTRATION

FastAPINext.jsLangChainLlamaIndexLangGraphCrewAI

DATABASES & VECTOR STORES

PostgreSQLRedisChromaDBPineconeFAISSQdrantLanceDB

CLOUD & INFRA

AWSAWS SageMakerDocker

01

AI Infrastructure

Custom-built systems from the ground up. No abstractions, no shortcuts.

~69ms @ 50K VECTORS

Hermes

A hybrid vector database built in Go + C++. Go handles the HTTP/concurrency layer; C++ powers Euclidean distance calculations via CGO. K-NN search on 1536-dim vectors with sub-100ms latency.

GoC++CGOREST API
Open case study
99.38% ACCURACY

Multi-Cancer Early Detection

Hybrid ResNet50 + Linear SVM pipeline for 5-class histopathology classification on the LC25000 dataset. Grad-CAM on layer4 for interpretability.

PyTorchResNet50SVMGrad-CAM
Open case study

02

Agentic Systems

Agents with memory, planning, and evals. Built to survive production.

MULTI-AGENT ORCHESTRATION

Agentic Finance Engine

A multi-agent financial analysis system using LangGraph. Orchestrates specialized agents for market data retrieval, portfolio analysis, and risk assessment with tool-use and memory.

LangGraphOpenAIChromaDBPython
Open case study
END-TO-END COMPLAINT MGMT

Aegis

A full-stack agentic customer complaint system with autonomous refund processing, reorder handling, and Human-in-the-Loop (HITL) escalation for edge cases.

LangGraphHITLAgentsPython
In Progress

03

RAG Projects

Retrieval systems that actually retrieve the right thing.

DOMAIN-SPECIFIC RETRIEVAL

Financial RAG

A RAG pipeline for financial document Q&A - chunking, embedding, and retrieval tuned for dense financial text with source-grounded answers.

ChromaDBOpenAIPythonLangChain
Open case study

[ NEXT BUILD ]

04

Research / Blogs

Writing that documents applied lessons from shipping AI systems, not just prototypes.

View all posts →

05

System Design

Current focus areas I use to architect dependable ML-native platforms.

Low-latency agent orchestration under failure conditions
Event-driven ingestion pipelines for RAG freshness
Cost-aware inference routing across multiple LLM providers
Observable AI systems with eval telemetry + feedback loops

Start a conversation

Need an AI system that is fast, observable, and safe in production?

I debug at the transformer level, build at the systems level, and ship at the production level.

Most people pick one. I got bored.

If your AI needs to actually work - not just demo well - I'm your person.

$ Available for freelance projects, consulting, and full-time roles.