01 / Index — AI Engineer — Personal Engineering Library

Rishav Raj

Computer Science Engineer with an AI focus — give me a problem, and I’ll find a way through it.

RAG · Agentic AI · Applied ML · Reliable AI Systems

Intelligent systems, engineered end to end — retrieval-grounded generation, governed agents, and applied ML held to executed baselines.

Computer Science Engineering student specializing in Artificial Intelligence — Generative AI, Agentic AI, and AI-native full-stack engineering.

I build practical software systems that combine artificial intelligence, data, and thoughtful engineering.

Explore Engineering LibraryAbout Me
Portrait of Rishav Raj
Rishav Raj — Portrait01

03 / Engineering Library

Systems, experiments, and engineering work.

A curated index of built artifacts. Each record carries an accession ID, a technology line, and its mapping onto the intelligence pipeline above — open a record to read the full dossier: problem, architecture, decisions, and evidence.

Real systems, honestly documented — every claim traceable to its repository.

07 records · 07 verified · 04 collections

Collections

Featured artifacts Projects

Meridian Market storefront — hero, four storefronts, category filters, and the 12-item catalog
FIG. 01 — The storefront: four vendor storefronts over a 12-item catalog, category filters, and the Ask Aria assistant in the header — committed capture shipped with the repository.

RR-SYS-005AI-Native Commerce / Transactional Full-Stack

VERIFIED

Meridian Market

Multi-vendor marketplace with a transactional core — atomic inventory reservation, integer-cents commission splits, an order state machine whose PAID transition belongs exclusively to an HMAC-verified payment webhook, a socket.io realtime layer with HMAC room tickets, and a catalog-grounded AI shopping assistant with a deterministic fallback.

Next.js 16 · TypeScript · React 19 · Prisma · SQLite · Socket.IO · Tailwind CSS 4 · Zustand · TanStack Query · Vitest

06 figures — full set in dossier

GitHub— github.com/rishav579/meridian-market (opens in a new tab)API / Docs— DECISIONS.md — twelve architecture decision records (opens in a new tab)
Secure Enterprise RAG documents manager on the author's Render-hosted deployment — tenant scope default under the ADMIN role, with the two ingested test PDFs listed at Employee access tier and COMPLETED ingestion status
FIG. 01 — The /documents surface of the author's own Render-hosted deployment, checked hands-on on 05 Sep 2026: tenant scope default under the ADMIN role, and the two ingested test PDFs at Employee access tier with COMPLETED ingestion status. Interface and ingestion state only — no retrieval or generation behavior is claimed from this capture.

RR-SYS-006AI Security / Multi-Tenant Retrieval

VERIFIED

Secure Enterprise RAG

Security-first multi-tenant RAG platform — tenant and document authorization enforced inside the retrieval SQL itself, hybrid pgvector + PostgreSQL full-text search fused by Reciprocal Rank Fusion with a local FlashRank cross-encoder, and grounded generation with PII scrubbing, prompt-injection defenses, citation verification, and explicit refusal behavior.

FastAPI · React 19 · TypeScript · PostgreSQL 16 · pgvector · FlashRank · Gemini

01 figures — full set in dossier

GitHub— github.com/rishav579/secure-enterprise-rag (opens in a new tab)Live Demo— Live deployment — Render (opens in a new tab — verified live deployment)API / Docs— API endpoint — Render (opens in a new tab)
Bhashini Voice Gateway demo dashboard — recognized command, structured action proposal, and human confirmation step
FIG. 01 — Demo dashboard from a locally executed run: the recognized code-mixed command, the structured action proposal, and the human confirmation step. Seeded demo data throughout; no production deployment and no live Bhashini credentials involved.

RR-SYS-007Multilingual Voice AI / Transaction Safety

VERIFIED

Bhashini Voice Gateway

Multilingual voice-to-action gateway — code-mixed Hindi/English voice commands become structured, validated, human-confirmed database actions behind a confirmation-token boundary, with transactional row locking, idempotency, and replay protection so a misheard sentence cannot move money without explicit user confirmation.

FastAPI · LangGraph · Pydantic · PostgreSQL · React 18 · TypeScript · Docker

01 figures — full set in dossier

GitHub— github.com/rishav579/bhashini-voice-gateway (opens in a new tab)
OWNARA workspace dashboard — morning brief, Kavya's current status, estimated recovery, and quick actions
FIG. 01 — Workspace dashboard right after demo provisioning: morning brief, Kavya's current status, estimated recovery across overdue invoices, and quick actions. Live deployment, seeded demo data.

RR-SYS-002Agentic AI / Business Operations

VERIFIED

OWNARA-AI

Governed AI execution system — AI operators run a closed Observe → Reason → Approve → Execute → Audit loop under persistent mandates with explicit authority boundaries, hashed Execution Contracts behind a human approval gate, and a hash-chained audit ledger. First operator: Kavya, an AI accounts-receivable operator.

Next.js 16 · React 19 · PostgreSQL · Prisma · LLM gateway

07 figures — full set in dossier

GitHub— github.com/rishav579/OWNARA-AI (opens in a new tab)Live Demo— Live deployment — Railway (opens in a new tab — verified live deployment)API / Docs— docs/RAILWAY-DEPLOYMENT.md (opens in a new tab)
Investigation workspace from an executed end-to-end run — status cards, citation verification, executive summary and root cause panel
FIG. 01 — Investigation workspace from an executed end-to-end run: 9/9 steps completed, 9 hashed artifacts, 33 lifecycle events, and the synthesized report passing citation validation. The repository's current documented default uses its deterministic offline LLM provider.

RR-SYS-004Agentic AI / Evidence Systems

VERIFIED

Enterprise AI Investigation

Evidence-grounded agentic investigation system — a deterministic planner dispatches read-only SQL and document tools, every artifact is SHA-256 hashed, and synthesis is a citation-validated report that refuses to name a root cause without evidence.

FastAPI · SQLAlchemy · React 19 · Vite · Pydantic · Docker

06 figures — full set in dossier

GitHub— github.com/rishav579/enterprise-ai-investigation (opens in a new tab)Live Demo— Live deployment — Railway (opens in a new tab — verified live deployment)API / Docs— docs/ARCHITECTURE.md (opens in a new tab)
Executive overview tab — 75 node positions, 26 critical stockout risks, $5,149.75 capital at risk, 48 open actions, and risk/action breakdown charts
FIG. 01 — Executive overview from an executed decision-intelligence run: 75 node positions, 26 critical stockout risks, $5,149.75 capital at risk, and 48 prescriptive actions (24 urgent).

RR-SYS-003Demand Forecasting / Decision Intelligence

VERIFIED

Demand-Risk Intelligence Engine

Forecast-aware supply-chain decision engine — a 40-check quality gate feeds relational marts, a LightGBM multi-horizon forecaster benchmarked against statistical baselines, a daily runout simulation with tiered safety-stock logic, and a prescriptive replenishment planner served over FastAPI and a Streamlit decision dashboard.

Python · LightGBM · pandas · SQLAlchemy · FastAPI · Streamlit

06 figures — full set in dossier

GitHub— github.com/rishav579/Real-Time-Demand-Risk-Intelligence-Engine (opens in a new tab)API / Docs— docs/ARCHITECTURE.md (opens in a new tab)
RepoPilot v0.1.0 web console: repository registration panel with rishav579/secure-enterprise-rag READY at 129 files / 536 chunks, the Grounded Answer panel citing five source blocks with pipeline telemetry, and five valid evidence cards listing file paths, line ranges, and symbols
FIG. 01 — The live deployed console (v0.1.0): a repository registered by HTTPS URL sits READY at 129 files / 536 chunks; the answer panel returns a grounded response over cited source blocks with pipeline telemetry (retrieval 87.9 ms, 8 candidates), and five evidence cards carry file paths, line ranges, symbols, and validity checks.

RR-SYS-001AI / Developer Tooling

VERIFIED

RepoPilot

Grounded repository Q&A engine — parses codebases into AST symbol trees, indexes them for hybrid retrieval, and answers natural-language questions with citation-verified evidence.

FastAPI · Tree-sitter · SQLite FTS5 · RAG

02 figures — full set in dossier

GitHub— github.com/rishav579/repo-pilot (opens in a new tab)Live Demo— Live deployment — Render (opens in a new tab — verified live deployment)API / Docs— API endpoint — Render (opens in a new tab)

Full holdings index Every record

Showing all 7 records

Verified — record audited against its linked repositoryReserved — slot held for a future artifact

04 / AI Engineering

Primary direction

AI Engineer.

The six systems above are one practice: I take AI products from idea to working software — answers grounded in retrieved sources, agents that stay within approved actions, and model results checked against baselines.

The Python backends and AI-native full-stack products around them are part of the same work — built, tested, and documented like everything else in the record.

Read from the record — not asserted ahead of it.

Evidence — verified records

Role directions

  1. Generative AI / LLM Engineer

    LLM features that answer from real sources — context retrieved into the prompt, citations checked before display, and honest fallbacks when the sources don't support an answer.

  2. Applied AI Engineer

    Machine learning for real decisions — data split by time to prevent leakage, results compared against simple baselines, and evaluation numbers published with the work.

  3. Agentic AI Engineer

    AI agents that use tools safely — every action approved and logged, with hard limits that keep autonomy under control.

  4. AI Software Engineer

    The engineering around the models — typed APIs, dependable state handling, authentication, realtime updates, and tests that actually run.

Capabilities

Every reference scrolls to its verified record in the Engineering Library — the dossier opens from the record itself.

05 / The Archivist

The person behind the record.

I'm Rishav Raj — a Computer Science Engineering candidate from Patna, Bihar, India, specializing in Artificial Intelligence. I graduate in 2026.

My focus is AI-native engineering — generative and LLM applications, retrieval-augmented systems, agentic and tool-using workflows, and the Python backends and data foundations that make them work in practice.

I like taking an idea all the way to working software.

Subject headings

  • Artificial Intelligence
  • Generative AI / LLM applications
  • RAG and retrieval systems
  • Agentic AI and tool-using workflows
  • Applied Machine Learning
  • AI-native full-stack development
  • Python / backend engineering
  • Data and intelligent systems
Portrait of Rishav Raj
Rishav Raj — The Archivist02Patna, Bihar, India · B.Tech CSE (AI) · Class of 2026

06 / Correspondence

Direct channels.

I read every message.

Emails17992307@gmail.comGitHubgithub.com/rishav579LinkedInlinkedin.com/in/rishav-raj-091524322/ResumeMaster CV — AI Engineer (PDF)