Medhansh Kapoor

Building a standalone dataset-quality evaluation toolkit for AIKosh (under MeitY's IndiaAI Mission) — bridging the ICMR MIDAS 2.0 framework gap, slated for production integration by internship end.

Next.js, FastAPI, Celery, Redis, Supabase (PostgreSQL), AWS S3, Pydantic, JWT Auth, Jinja2, WeasyPrint

• Metadata Intake: Engineered an 8-step wizard with pre-signed URLs for direct-to-storage uploads, bypassing the backend to enhance security and reduce server load.
• Assessment Engine: Architected an asynchronous, parallelized scoring engine to evaluate datasets across 15 government-defined quality domains, generating Composite Quality and Privacy Risk scores for automated release eligibility (Open/Controlled/Restricted).
• Integration: Built automated multi-format reporting (JSON/HTML/PDF) and webhook APIs, enabling AIKosh to auto-ingest verified dataset metadata upon assessment completion.

Building a self-contained, fully offline GIS platform enabling classified defence environments with no internet access to securely upload, process, and visualize geospatial map data.

FastAPI, PostGIS, GDAL/ogr2ogr, GeoServer, Martin Tile Server, OpenLayers, Docker/Podman, RHEL

• Ingestion Pipelines: Built automated ETL pipelines for vector (Shapefiles to PostGIS with reprojection/indexing) and raster data (GeoTIFFs auto-published via REST), eliminating manual GIS server setups.
• Serving Layer & Infra: Enabled real-time map-tile delivery to browser clients via a containerized 4-service microservices backend, securely deployed on a firewalled RHEL environment for fully offline, classified operations.

Evaluated enterprise AI toolchains and shipped production integration tests on live Salesforce CRM infrastructure.

Vanna.ai, LlamaIndex, LangChain, Librosa, PyDub, Torchaudio, Salesforce Apex, Einstein AI, Agentforce

• Benchmarking: Evaluated Vanna.ai, LlamaIndex, LangChain for NL-to-SQL, and audio frameworks (Librosa, PyDub) for feature extraction; identified that agent architecture outweighs model choice for multi-table reasoning.
• Salesforce AI Platform: Built Apex REST services and SOAP integrations; conducted prompt engineering experiments on live CRM data using Einstein AI and Agentforce.

Detects civic issues via CCTV/dashcam, auto-generates complaint tickets, verifies repairs, and retrains itself.

YOLOv8, ONNX, OpenCV, FastAPI, PostgreSQL/PostGIS, Gemini 2.5

• Model & Inference: Trained YOLOv8 on 4,783 images with 30% engineered negatives and FN-bucketing for recall diagnosis; achieved 20ms ONNX inference on a FastAPI microservice deployed via GCP Cloud Run.
• Auto-Ticketing & Reliability: Engineered a burst-frame extraction pipeline feeding a 4-tier reliability engine and DIGIPIN geospatial deduplication (4m² grid), processing 256 end-to-end complaints in 0.36s per ticket.
• Verification & Active Learning: Enforced automated post-repair rescans to verify fixes before ticket closure; bucketed field data into labeled classes for continuous baseline-gated retraining.
• LLM Routing (Seva): Integrated Gemini 2.5 Flash using a 4-step Chain-of-Thought to autonomously route complaints across 42 civic categories, auto-routing 125/256 tickets with zero manual input.

5-agent pipeline: legal case in — research, strategy, drafted documents, and reasoned explainability out.

LangGraph, FastAPI, Qdrant, Neo4j, Redis, WebSockets, Celery, Groq, Gemini

• Orchestration & Intake: Built a stateful 5-stage LangGraph pipeline (Intake → Research → Strategy → Drafting → Explainability) with OCR-aware processing and SHA-256 caching to mitigate LLM hallucinations.
• RAG & Determinism: Indexed 4,582 chunks (7 legal acts) in Qdrant via Sentence Transformers; engineered a diagnostic harness to catch retrieval drift and OCR instability across N-run tests.
• Reliability & Graph: Implemented a triple-engine fallback (Groq → Gemini → offline rules) for zero-downtime generation; built a Neo4j legal knowledge graph (1,410 nodes, 1,837 relationships) for explainable document drafting.
• Realtime Infrastructure: Developed an event-driven custom Redis Pub/Sub, WebSockets, and Celery backend for live citizen-lawyer negotiation chat, bypassing third-party APIs.

India Innovates '26 — National Finalist

Winner in Digital Democracy track; presented JanSamadhan live before senior policy leaders at Bharat Mandapam; project forwarded to central ministries.

Prayatna 3.0 Hackathon — Finalist

Built NyayaAI's 5-agent legal AI backend during a 36-hour sprint.