Bhubaneswar, India · Open to AI/ML & Computer Vision roles

Satyajit Parida — building AI systems that actually ship.

Final-year CSE student. IEEE-published researcher. I design agentic RAG pipelines, real-time MLOps loops, and edge-deployed computer-vision systems.

Open to internships & full-time roles
9.75/10
SGPA
0
Public Repos
0
IEEE Paper
0
Day Streak
0
Job Simulations
IEEE Member #101009704
SP

About

The person behind the systems.

Final-year B.Tech CSE, Centurion University, Bhubaneswar (2022–2026), SGPA 9.75/10. Focused on systems that hold up outside a notebook.

Builds production-grade agentic RAG pipelines, MLOps loops, multi-agent LLM apps, edge-deployed CV systems. Published at IEEE CCPIS 2025.

Cares about systems that survive contact with the real world — clean code, reproducible pipelines, honest metrics.

Location

Bhubaneswar, Odisha

Education

B.Tech CSE, 2026

Focus

AI/ML · CV · MLOps

Status

Open to roles

Selected Work

Nine systems worth a closer look.

Click any card for the full architecture breakdown.

01/09Cybersecurity

Invisible Shield

Behavioral biometrics bot detection — mouse velocity, keystroke rhythm, IEEE-published.

  • IEEE CCPIS 2025 published
  • Velocity + jerk signal analysis
  • 100% offline, zero cloud APIs
Pythonscikit-learnFeature EngineeringFastAPI
02/09InsurTech

ClaimPilot

6-agent LangGraph state machine — document to settlement PDF in 2.4 seconds.

  • 2.4s avg per claim
  • 6 specialized agents, conditional edges
  • Human escalation on ambiguous cases
LangGraphPythonFastAPIPostgreSQL
03/09MLOps

FraudStream

UPI fraud detection — XGBoost, SHAP explainability, automatic drift-triggered retraining.

  • <80ms scoring latency
  • 15 Docker services
  • Auto-retrain on 10% drift
XGBoostSHAPDocker ComposeRedisEvidently
04/09LegalTech

ClauseGuard

7-stage contract intelligence pipeline — clause risk scoring, RAG Q&A, zero hallucination tolerance.

  • 30 clause types extracted
  • Temp=0 + hardcoded refusal phrase
  • Per-contract Pinecone namespaces
LangChainPineconeOpenAIFastAPIReact
05/09RegTech

RegWatch

Autonomous regulatory monitoring — scrapes, classifies, alerts. Human sign-off before any action.

  • 4+ regulator sources scraped
  • Risk-tiered Slack/email alerts
  • Human-in-the-loop by design
PythonBeautifulSoupCeleryPostgreSQL
06/09Computer Vision

SentryLens

Real-time PPE violation detection over RTSP — YOLOv8, SMS alert within 15 seconds.

  • ~10fps effective inference
  • 3–15s alert latency
  • Zone-polygon violation mapping
YOLOv8OpenCVPythonTwilio
07/09Customer Support AI

SupportForge

Stateful multi-agent support system — 21 enforced budget invariants, fast-path for 30% of tickets.

  • 21 budget invariants enforced
  • 30–40% fast-path (no LLM call)
  • BM25 + vector hybrid retrieval
LangGraphBM25pgvectorFastAPI
08/09GovTech / Mobile

RoadSaathi

Offline-first Android hazard reporting — on-device TFLite inference, PostGIS geofencing.

  • 87% on-device classification accuracy
  • ~200ms inference
  • 3-tier: Android/Spring/React
AndroidTFLiteSpring BootPostGISReact
09/09Self-Hosted RAG

DocRAG

8-node LangGraph RAG for business PDFs — fully local, Ollama + ChromaDB, zero data leaves the machine.

  • 8-node agentic pipeline
  • 100% self-hosted, no API calls
  • Page-cited, grounded answers only
LangGraphOllamaChromaDBLlama 3.3

The Archive

28 more, shipped along the way.

Smaller in scope. Still real code, not tutorials.

Link visibility is checked against github.com/sat1828 — a card only shows a GitHub link if the repo is confirmed public.

Toolbox

What I reach for, organised by intent.

Click any skill to see what it was actually used to build.

{ }

Languages

AI / ML

Agents & RAG

MLOps / DevOps

Backend / Web

Data & Cloud

< Soft_Skills />

Beyond the code.

Hover or focus each tile to see the proof.

Problem Solving

hover to reveal →

Problem Solving

Cut FraudStream's scoring latency from ~340ms to under 80ms by tracing the bottleneck to redundant feature recomputation, not model complexity.

Critical Thinking

hover to reveal →

Critical Thinking

Chose XGBoost over a deep model for FraudStream specifically for SHAP-level explainability and sub-80ms latency — accuracy alone wasn't the deciding metric.

Teamwork

hover to reveal →

Teamwork

Coordinated with IEEE CCPIS reviewers across two revision rounds to align Invisible Shield's methodology section with peer feedback.

Communication

hover to reveal →

Communication

Distilled a keystroke/mouse-dynamics detection pipeline into a peer-reviewed IEEE paper a non-specialist committee could evaluate.

Adaptability

hover to reveal →

Adaptability

Picked up LangGraph from scratch in under three weeks to rebuild ClaimPilot's pipeline as a conditional-edge state machine.

{ }

Independence

hover to reveal →

Independence

Owned SupportForge end-to-end — architecture, all 21 budget invariants, and deployment — as a solo build.

Org. Skills

hover to reveal →

Org. Skills

Shipped ClauseGuard and RegWatch in the same semester as a 9.75 SGPA and three AWS certifications.

Conscientiousness

hover to reveal →

Conscientiousness

Maintains a live, re-verifiable GitHub contribution streak — check github.com/sat1828 for the current count.

Practical Intelligence

hover to reveal →

Practical Intelligence

Chose on-device TFLite inference for RoadSaathi specifically because rural connectivity couldn't be assumed — 87% accuracy at ~200ms, offline-first by design.

Experience

Time on the job.

ML/AI Intern

· Uptoskills
RemoteJan – Apr 2025
  • Built MCQ generation system (FastAPI, 1,000+ questions)
  • Raised endpoint test coverage from 0% to 80%
  • Delivered content evaluation module for a full-stack assessment platform

Recognition

Proof, not adjectives.

IEEE CCPIS 2025

SGPA 9.75/10

TechNova Finalist

Top-25 GfG Hackathon

100

100-day problem-solving streak — measured discipline, not a highlight reel.

Certifications

OR

Oracle

OCI Gen AI Professional

Aug 2025

OR

Oracle

OCI AI Foundations Associate

2025

GO

Google Cloud

GenAI Leader Path — 5 courses

2025

AW

AWS

Cloud Practitioner

2024

AW

AWS

Gen AI: Art of the Possible

Oct 2025

AW

AWS

DevOps on AWS

Oct 2025

AW

AWS

ML Specialty Exam Prep

Oct 2025

HA

HackerRank

Certified Software Engineer

2026

KA

Kaggle

Python for Data Science & AI

2025

GE

GeeksforGeeks

DSA w/ Centurion University

Writing & Research

One paper. Real, peer-reviewed.

IEEE CCPIS 2025

Invisible-Shield: Behavioral Bot Detection via Keystroke and Mouse Dynamics

The paper proposes a behavioral biometrics approach to bot detection that models the physics of human mouse motion and keystroke timing rather than relying on request-level signals alone. Evaluated against simulated and real interaction traces, it demonstrates that motion-dynamics features carry enough signal to separate human sessions from scripted ones without any client-side fingerprinting.

Permanent IEEE Xplore link pending indexing — link will appear here once available.

IEEE Member #101009704

More write-ups on RAG architecture and MLOps in progress — no fixed date.

Let's build something worth shipping.

Open to AI/ML, Computer Vision, and MLOps roles — internships and full-time.

Email

satyajitparida294@gmail.com

Send email

Phone

+91 8260743307

Call

Schedule

Quick intro call, no pressure