Define state, inputs, constraints, and failure paths before choosing the model or framework.
Atishay Jain / Computer Engineering
AI systems. Product workflows. Proof-first engineering.
AI Systems Engineer Product Engineer
I build AI-enabled products by turning models, APIs, data, and user flows into systems with explicit fallbacks, clear workflows, and honest limits.
How the work changed my engineering
I learned to design the whole decision path.
My projects moved me from isolated algorithms and models toward systems thinking: start with contracts, route uncertainty deliberately, persist the state that matters, and treat the user workflow as part of the architecture.
Computer Engineering / software systems / local AI / product workflows / practical automation
Use rules, local models, memory, and LLM fallback only where each earns its complexity.
Connect backend decisions to interfaces, feedback, recovery, and inspectable proof.
Featured projects
Three systems that changed how I build
Adaptive behavior, local AI routing, and campus coordination each introduced a different engineering constraint.
FEATURED / PROTOTYPE
Mahoraga
Adaptive enemy behavior shaped through environment rules and reward signals.
Reward design is system design: environment rules, pressure, and failure modes shape behavior as much as the model.
FEATURED / LOCAL AI SYSTEM
Hybrid GenAI Transaction Categorizer
Categorize transactions without making the LLM the default path.
Uncertainty should be routed, not handed to the largest model by default.
FEATURED / PRODUCT SYSTEM
The Loop
Connect event discovery, interest context, RSVP, social coordination, chat, and carpool planning.
A product becomes useful when discovery connects to persisted coordination workflows.
Signal evolution
How the approach evolved
Each stage added a new constraint: algorithms, complete workflows, routed uncertainty, adaptive behavior, then production discipline.
01
C++/DSA / ML basics / SQL foundations
Learned to reason from algorithms, data shape, and query structure before reaching for frameworks.
02
The Loop
Learned that useful products connect discovery to persisted coordination, not isolated screens.
03
Hybrid GenAI Transaction Categorizer
Learned to route ambiguity through rules, local inference, memory, and fallback paths.
04
Mahoraga
Learned that environments, reward signals, and evaluation limits shape agent behavior.
05
ReceiptSplit / AI workflow systems
Applying those lessons to clearer contracts, validation, and production-minded workflows.
Toolkit
Tools chosen for the job
Grouped from project evidence; the stack supports the engineering decisions above.
Python TypeScript JavaScript C++ SQL
React Next.js Astro Vite Tailwind CSS shadcn/ui
FastAPI Node.js REST APIs WebSockets Uvicorn
ONNX Runtime DistilBERT Qwen2.5 GGUF llama.cpp PyTorch Gymnasium LoRA
PostgreSQL SQLite Supabase SQLAlchemy
Docker GitHub Actions GitHub Render
Auth/JWT Realtime OCR pipelines UPI deep links Maps Recommendations
Systems map
Capabilities traced back to projects
The recurring patterns across AI, product, data, and automation work—after the proof, not instead of it.
AI systems architecture
Mahoraga / Hybrid Categorizer / AI Resume Tailoring Engine
Product engineering
The Loop / ReceiptSplit
Automation
Resume Engine / Webcam Motion Alert / Music Web Scraper
Data and ML foundations
Minimal RAG / Global Super Store / Fast and Curious / Linear Regression
Full-stack implementation
The Loop / Hybrid Categorizer / ReceiptSplit
C++/DSA foundations
Coursework and resume signal
Secondary work
Smaller systems, practical constraints
Automation builds that extend the main story without competing with the featured work.
SECONDARY / REPO VERIFIED, COMPACT SUPPORT
AI Resume Tailoring Engine
Constrained LLM workflow automation for resume generation.
Automate resume tailoring while avoiding unsupported claims and private-data leakage.
SECONDARY / REPO VERIFIED, COMPACT SUPPORT
Webcam Motion Alert System
OpenCV and alerting workflow for motion-triggered notifications.
Detect motion and trigger useful alerts with cleanup instead of a passive camera script.
Lab Projects
Experiments that built the foundations
Retrieval, scraping, database, and analysis work kept compact and linked to source.
Minimal RAG Implementation
Retrieval, chunking, embeddings, and grounded-answer workflow.
Focused retrieval experiment.
Music Web Scraper
Scraping pipeline for collecting and structuring music data.
Practical data extraction build.
Global Super Store Database Project
SQL schema design, queries, and dashboard-ready business data.
Database fundamentals proof.
Fast and Curious Car Analysis
Exploratory car dataset analysis with visual insights.
EDA and visualization exercise.
Contact
Inspect the evidence, then let's talk.
For software, AI systems, or product engineering opportunities, these are the direct paths.