Rashanjot Kaur | Agentic AI · Published Researcher

// SECTION: PROJECTS

001

════ LIVE IN PRODUCTION ════

terrier.grader · langgraph.dag nodes:4

HW6

↓

RUBRIC

↓

SOLN

↓

GRADE×50

Terrier Grader

Multi-Agent Academic Evaluation Pipeline

Boston University · Prof. Eugene Pinsky

■ PRODUCTION

Aug 2025 – Present

RAGAS EVAL (LangSmith):

faithfulness0.97

relevancy0.94

correctness0.91

3BU COURSES

50+STUDENTS/SEM

60%OVERHEAD REDUCED

150+DOCS/CYCLE

LangGraph · Bedrock · Lambda · DynamoDB · S3 · CDK · CI/CD

Skills: ✓ Multi-agent (LangGraph) · ✓ LLM Evals (RAGAS+LangSmith) · ✓ Production

BizSight

Multi-Tenant Business Intelligence AI Platform on Bedrock AgentCore

■ PRODUCTION

MULTI-TENANT INPUT

Tenant Atenant1_biz1

Tenant Btenant2_biz1

Tenant Ctenant3_biz1

JWT → tenant_idper-tenant S3

▼

INFRAECR · IAM · S3 · DynamoDB

container registry · roles · per-tenant buckets · metadata

│ provisions ↓

AUTHCognito · JWT → tenant_id

user pool · sign-in · token validation · multi-tenant isolation

│ authenticated request ↓

RUNTIMEBedrock AgentCore · Docker

ORCHESTRATORroutes requestssession mgmt

INSIGHTStab_pulsetab_journeytab_actions

SUPPORTchatfeedback

← 3 Strands agents

│ agent response ↓

APIGateway · Lambda proxy

/api/onboard · /api/tabs · /api/chat · /invocations

│ JSON response ↓

FRONTENDCloudFront · S3 · React · TypeScript · shadcn/ui

Dashboard · Chat · Onboarding UI

▼

tenant-scoped dashboard · real-time insights · chat support

5 CDK stacks · 3 agents · AgentCore · Strands · React · CDK

Skills: ✓ AWS depth (AgentCore+CDK+Cognito) · ✓ Multi-tenant SaaS · ✓ Full-stack

SWAAD

MCP-extensible restaurant recommendation with 6-dim taste vectors

■ PRODUCTION

FLAVOR PROFILE:

sweet██░░░░0.3

umami█████░0.8

salty███░░░0.5

sour█░░░░░0.2

spicy████░░0.7

bitter██░░░░0.4

TASTE_VECTOR

{umami:0.8}

──●──

4 AGENTS

Yelp · Flavor · Bev · Budget

──●──

MATCH + MCP

semantic search

Strands · Bedrock · Yelp API · 6-Dim Taste Vectors · MCP

Skills: ✓ MCP (find_restaurant) · ✓ Semantic search · ✓ Multi-agent (Strands)

════ ACTIVE DEVELOPMENT ════

Mindwell

DPO-Aligned Mental Health AI with Clinical Safety Scoring

○ ACTIVE DEVELOPMENT

INPUT

HIDDEN

OUTPUT

DPO fine-tuned · Llama 3.2 1B

DPO TRAINING: 82%

pairs: 4.2K · epochs: 3 · lr: 2e-5 · β: 0.1

PHQ-912/27moderate

GAD-714/21moderate

C-SSRS0/6no risk

■ Journal entry → NLP analysis ─────────────

■ PHQ-9 / GAD-7 / C-SSRS scoring ─────────

■ Bedrock Guardrails safety layer ────────

● Risk flag + personalized response

Strands · FastAPI · Bedrock · HuggingFace

Skills: ✓ LLM Fine-tuning (DPO) · ✓ Guardrails (Bedrock) · ✓ Healthcare AI

sprint.algo.statusSTATUS: DEV

SprintAgent

Personality-Aware Sprint Planning with MCP

○ ACTIVE DEVELOPMENT

■ Slack / Taiga task input ─────────────────

■ SBERT embed → XGBoost story points ──────

■ Hungarian cost matrix × personality ─────

● MCP: assign_task() → Taiga

MCP Coverage● READY

AlgorithmPurposeStatusComplexity

SBERTStory point embed○ DEV384-dim

HUNGARIANTask assignment○ DEVO(n³)

KNAPSACKSprint planning○ DEVO(n·W)

TOPO_SORTDependency resolve○ DEVO(V+E)

BIG FIVE → TASK ROUTING

CON0.89→ deadline-critical

EXT0.72→ collaborative

OPN0.65→ exploratory

AGR0.91→ conflict resolution

NEU0.31→ high-stress

O(n³) HUNGARIAN508K TRAININGMCP SERVER<100ms ASSIGN

Strands · SBERT · XGBoost · PostgreSQL · MCP · Taiga API

Skills: ✓ MCP · ✓ Optimization (Hungarian+Knapsack+TopoSort) · ✓ ML pipeline

════ RESEARCH ════

GridShift

Agentic Carbon-Aware Scheduling of AI Workloads

○ RESEARCH

CCI Framework published: Energies 2026

ISO-NE 24h CARBON

HR0003060912151821

CI█░░████░

HILOLOHIHIHIHILO

SCHEDTRTR──IF────TR──

■ Forecast Agent · Electricity Maps 72hr

■ CCI Agent · 1× infer → 4.3× train

○ Scheduler Agent · greedy / MILP

○ Orchestrator · coordinates

Extends: CCI Framework · DOI: 10.3390/en19030642

Electricity Maps · greedy → MILP

Strands · Bedrock · Electricity Maps · CCI · MILP

Skills: ✓ Multi-agent · ✓ Carbon-aware AI · ✓ Research→Production

MESA

Multi-Agent Ensemble for Stock Analysis

○ RESEARCH

Targeting publication 2026

SIGNALS → AGENTS → OUTPUT

ETF CORR ≥0.99OHLCV yfinanceSENTIMENT 5 models

──●──

9 AGENTSetf_analysis · data_preprocess · feature_eng · lstm · cnn · rnn · ensemble_coord · backtest · orchestrator

──●──

BUY / HOLD / SELL+ confidence

WHY ETF?

AAPL · XLK · QQQ → ≥0.99 correlation. XLK holds AAPL → move together → use as input feature.

LSTM+CNN+RNN · 25yr backtest (2000–2024) · 5-model sentiment fusion

PyTorch · Strands · Bedrock · scikit-learn · yfinance

Skills: ✓ ML ensemble · ✓ Multi-agent (9+ Strands) · ✓ Sentiment analysis

✉ Interested in what I build?contact@anacodic.ai →