Hi, I'm Ashwin Mishra
I build LLM agents, RAG pipelines and computer vision systems that ship: grievance platforms for government portals, AI study tools for UPSC aspirants, legal-document intelligence, and agentic MarTech tools for brands like Hyundai and Kajaria. Feel free to reach out :)
// currently into: agentic & multi-agent systems Β· retrieval-augmented generation Β· educational & Indic NLP Β· document intelligence & OCR Β· generative image pipelines
## Projects
/* Innocean India: INN agents & brand AI */
ScoutINNtelligence π
// Innocean India Β· 2026
Dual-intelligence lead generation. Supply Scan ranks startups by fit; Demand Scan qualifies buyers with a 3-filter check (MarTech need, buying capacity, actionability).
- FastAPI
- Next.js
- Groq
- Supabase
- Tavily
- SerpAPI
AdINNtelligence π
// Innocean India Β· 2026
Agentic competitive-ad intelligence for Hyundai's automotive marketing teams across 40+ brands. Collects competitor Meta ads, classifies models, offers and patterns with LLMs, and returns three evidence-backed recommendations for a dealer's next campaign.
- FastAPI
- React
- Groq
- Supabase
- Redis
- Meta Ad Library
Creative Intelligence Engine (Hyundai) private
// Innocean India Β· 2026
AI-generated, brand-safe ad creatives for Hyundai, built at Innocean India. Separate dashboards for admins, brand teams and dealers; campaign jobs queued on BullMQ.
- Next.js 15
- Express 5
- TypeScript
- BullMQ
- Prisma
- Claude
- fal.ai
- AWS
Kajaria Tile Visualizer private
// Innocean India Β· 2026
Full-stack tile visualization for Kajaria: upload a room photo, detect the floor, and overlay any of 60+ tiles. An auto-visualize mode uses AI room analysis to recommend tiles, with a custom tile upload, lead capture and a CRM drawer.
- Flask
- React
- Vite
- Claude
- fal.ai
/* IIT Kanpur */
ES-RAG Grievance Retrieval System π
// CDIS, IIT Kanpur Β· 2025
Plain-English search over a government grievance portal: an LLM turns the query into Elasticsearch DSL, a semantic cache skips repeat work, and hybrid keyword + vector retrieval feeds a RAG answer. Cut manual filtering and search time by ~50%.
- Elasticsearch
- Sentence Transformers
- RAG
- LLMs
- Python
Evidence Analysis π
// CDIS, IIT Kanpur Β· Kanpur Smart City Β· 2026
Evidence-image validation for Kanpur Smart City. Multi-factor scoring (variance, edge density, entropy, OCR layout) plus a voting pipeline flags documents that were uploaded where a site photo was expected.
- Python
- FastAPI
- React
- OCR
- Image processing
Radar Drone Classification private
// IIT Kanpur Β· 2024β25
Classifies drone type from multi-frequency (26β40 GHz) radar cross-section measurements interpolated with cubic splines across phiβtheta angles. 96.6% accuracy.
- Python
- SciPy
- YOLO
- Radar RCS
/* Independent */
Law Simplified π
// 2026
Legal document intelligence: multilingual OCR, clause and entity extraction, schema-driven summaries, and RAG chat grounded in the uploaded document. Improved case-review efficiency by 60%.
- OCR
- Transformers
- RAG
- Streamlit
## Explore more
[Resume (PDF)] [Case studies] [GitHub] [LinkedIn]
Open to AI/ML roles, research collaborations and interesting problems. Email me at ashwin.mishra07@gmail.com.