00.00.00
Projects
(click to view details)
ENT MonitorLiveGSoC
LangGraphPythonOpenAI
Pediatric health surveillance system tracking viral social media trends via an agentic LangGraph loop.
LangGraphPythonOpenAI
- •Architected an 8-node LangGraph agentic loop pairing SBERT + HDBSCAN clustering with a PubMed/Semantic Scholar cascade
- •Engineered continuous lifecycle tracking with dynamic velocity routing and early-warning signals instead of one-off alerts
- •Prevented medical hallucinations via deterministic evidence scoring and automated NCBI database cross-referencing
- •Consolidated APIs into Next.js on a hybrid GCP setup, cutting cloud infrastructure costs by 55%
PythonLangGraphOpenAIGCPNext.jsTSX
ShadowMapBuilding
PyTorchFastAPINext.js
On-device privacy extension that intercepts sensitive user data before it reaches external AI models.
PyTorchFastAPINext.js
- •Developing a Chrome extension that runs on-device NLP to detect and redact PII before it leaves the browser
- •Training a lightweight PyTorch transformer model for real-time named entity recognition on sensitive fields
- •Building a FastAPI gateway that applies differential privacy noise to structured data before forwarding to LLM APIs
- •Integrating Supabase for user preference storage and audit logging of all intercepted data events
PythonPyTorchFastAPINext.jsSupabaseChrome Extension
VerifyXLive
PythonFastAPIQdrant
XAI compliance platform that automatically verifies financial claims against regulatory RBI documents.
PythonFastAPIQdrant
500+
RBI Docs Indexed
10,000+
Chunks Indexed
94%
NLI Accuracy
96%
Top-5 Recall
- •Built a hybrid retrieval pipeline combining Qdrant vector search with Elasticsearch BM25 for maximum recall across RBI circulars
- •Integrated DeBERTa-based NLI models for explainable claim verification with confidence scoring and evidence highlighting
- •Designed a FastAPI backend with streaming responses and Supabase auth for real-time compliance dashboards
- •Engineered a document ingestion pipeline that chunks, embeds, and indexes regulatory PDFs automatically
PythonFastAPIQdrantElasticsearchGroqDeBERTaSupabase
HabitatLive
Next.jsXGBoostFastAPI
Predictive geospatial tool that maps species-specific wildlife habitat suitability from satellite imagery.
Next.jsXGBoostFastAPI
12
Species Mapped
32+
Districts Analyzed
4
Env Layers Modeled
89%
Accuracy
- •Built an end-to-end pipeline ingesting Sentinel-2 satellite imagery and extracting habitat features with NASA Prithvi foundation model
- •Trained scikit-learn ensemble classifiers across 4 environmental layers (rainfall, drought risk, temp, soil) for 32+ Maharashtra districts
- •Designed an architecture built for regional validation that generalizes to new geographic areas with regional observation data
- •Created an interactive map interface in Next.js with layer toggles for NDVI, land cover, and prediction overlays
PythonFastAPIscikit-learnSentinel-2PrithviSupabaseNext.js