Hi, my name is
Kishan K.
I build production AI systems.
I'm a Software Engineer (AI/ML) who builds and ships production GenAI, LLM, and computer-vision systems — from multimodal pipelines and RAG to cloud-native FastAPI backends on AWS.
About Me
Hello! I'm Kishan, a Software Engineer (AI/ML) with a strong foundation from Ramaiah Institute of Technology. I work across the full lifecycle — from research and modeling to deploying robust, scalable systems in production.
Currently at Silo Fortune, I build production AI for agriculture and livestock — computer-vision crop diagnostics, satellite intelligence (NDVI, EVI, NDWI), weather-driven recommendations, and multilingual LLM advisory generation across 12 Indian languages — on cloud-native FastAPI services with two-level template caching that reuses reports instead of regenerating them. Alongside this, I work part-time as a Forward Deployed Engineer for Narvia Design, a luxury interior/exterior design studio, where I built and shipped their production site end-to-end as sole engineer.
Education
Ramaiah Institute of Technology, Bengaluru
B.E. in Artificial Intelligence and Machine Learning
2021 - 2025 | CGPA: 8.6/10
Technical Skills
GenAI
AI/ML
LLM Engineering
MLOps & Cloud
Backend Engineering
Programming
Databases
APIs & Integrations
Experience
Software Engineer - AI/ML @ Silo Fortune
Oct 2025 - Present | Bengaluru
- Fine-tuned Qwen-2.5-1.5B (4-bit QLoRA) on 178,980 knowledge-base-generated advisory pairs covering 312 crops, 1,693 diseases and 12 Indian languages; on 300 validation prompts it produced the required 4-section advisory format 300/300 times (base model: 0/300), matching the action field 99.3% and the infection-rate range 85.7% of the time.
- Built the mandi price-forecasting service for 6,273 mandis and 346 commodities: a 6-model ensemble (CatBoost, LightGBM, XGBoost, GBR, RF, SVR) with lag and sin/cos seasonality features and a 12-week recursive forecast. A rolling-origin backtest on 15 market–commodity series (59 origins) gave 13.4% MAPE vs 17.0% for a naive baseline.
- Designed a two-level report cache (report template keyed on 5 bucketed satellite metrics, plus a per-language translation store in PostgreSQL): 99.4% template hit rate and about 97% fewer Gemini calls (8,130 hits vs 52 generations, production counters).
- Contributed to satellite processing (NDVI/EVI/NDWI) and an ISRIC soil-based crop-recommendation engine; services run on AWS ECS Fargate with Docker and Jenkins. The platform has reached 50,000+ installs and 3,500+ subscribers.
Forward Deployed Engineer @ Narvia Design
Aug 2026 - Present | Remote · Part-Time / Freelance
- Embedded directly with a luxury interior/exterior design studio client to design, build, and ship narviadesign.com end-to-end as sole engineer, from information architecture through production deployment. (References available on request.)
Applied ML Engineer @ Alpify Technologies LLP
Sep 2026 - Present | Bengaluru · Part-Time / Contract
- Contracted through Alpify to build a lead-quality pipeline for one of their clients: an ETL unifying 611 leads from 12 sheets across 6 channels (Meta, Google, outbound, website) and an XGBoost contactability model (5-fold CV AUC 0.81) that assigns P1/P2/P3 call priorities. Rule-based lead validation (phone, email and duplicate checks) is in progress. (References available on request.)
AI-Automation Intern @ Sumedha Design Systems
Feb 2025 - May 2025 | Hyderabad
- Engineered a scalable data-extraction pipeline using GCP Vision and AWS S3 to process 1,000+ VLSI documents, with RAG-based automated MCQ generation.
- Architected a hybrid retrieval framework with LlamaIndex and FAISS using re-ranking strategies — improving learner engagement by 15% and reducing latency by 50%.
Machine Learning Intern @ Pragami Solutions
Jan 2025 - Feb 2025 | Bengaluru
- Delivered a full-stack Productivity Management System to streamline operations for small-scale industries.
- Conducted research on lung disease prediction and interpretable models for healthcare datasets.
Things I've Built
Kisaan Sampurna
AI-Powered Agricultural Advisory Platform · @ Silo Fortune
Mandi Price Explorer
FastAPI backend serving price data for 6,273 agricultural markets across India. Real-time APMC market price discovery, automated recursive ML price forecasting, 12-language support, dual-source data (PostgreSQL + AWS S3 Parquet), and cascade dropdowns under 5ms via in-memory cache.
FastAPI, PostgreSQL, AWS S3, Docker, SQLAlchemy, JWT
Crop Stage Predictor
5-layer pipeline reading 5 years of satellite imagery (NDVI, EVI, NDMI) to detect crop growth cycles for 21 crops across 80+ sub-stages. Generates pest/disease advisories via parallel asyncio Gemini calls. Delivered in 10 Indian languages via cache-first translation.
FastAPI, Gemini 2.0 Flash, AgroMonitoring API, PostgreSQL, AWS S3
AI Farm Intelligence System
Satellite indices (NDVI, NRI, EVI, DSWI) → deterministic bucketing → 432 condition fingerprints → 15-section advisory in 13 languages. A two-level LLM cache reuses reports instead of regenerating them. 6 satellite overlay images stored on AWS S3. Deployed on AWS ECS Fargate + Jenkins CI/CD.
FastAPI, Gemini, AWS S3, AWS ECS Fargate, PostgreSQL, Jenkins
Crop Disease Scanner
Diagnostic system identifying 360+ crop varieties in 10 languages using Knowledge-Based Agents and Gemini Pro (API) models.
Python, Computer Vision, DL, ML Agents, Expert Systems
Kissan Sampurna Web
React Native frontend for agricultural ecosystem, enabling seamless crop image uploads and LLM-driven advisories with < 2s response times.
React Native, Node.js, Vite, AWS S3, Tailwind CSS
Agri-Expert Copilot
Proof-of-concept advisory copilot over a Neo4j knowledge graph built from a 3,140 crop–disease knowledge base and 19,000+ PIN-code zone mappings. A LangGraph flow (extract entities → traverse graph → synthesise advice) answers disease, dosage, phenology and crop-suitability queries, served via a Strawberry GraphQL API on FastAPI with RabbitMQ workers and an in-memory fallback graph.
Neo4j, Cypher, GraphQL, LangGraph, FastAPI, RabbitMQ
Sentinel
Agentic Research Copilot for Indian Equities · Personal Project
Multi-Agent Research Platform
Research-to-compliance agent for NSE/BSE-listed companies built with LangGraph: hybrid retrieval (BM25 + pgvector, reciprocal-rank fusion, BGE reranker) drafts a cited brief, then a QLoRA-tuned Qwen-2.5-7B audits it for SEBI rules (270/271 on a held-out synthetic set; a simple TF-IDF baseline reaches 269). CI with lint, type checks, 27 tests and a citation-validity checker.
LangGraph, QLoRA, Qwen 2.5, RAG, pgvector, RAGAS, FastAPI, PostgreSQL, Docker
NotifKit
Open-Source Notification Infrastructure · Personal / Open Source
NotifKit — Notification Infrastructure
Open-source notification infrastructure that provides a single API for routing notifications across email, WhatsApp, Telegram, Discord, SMS, push, Slack, and Teams. Handles intelligent routing, retries and fallbacks, templates, delivery analytics, event-triggered workflows, and optional AI-powered personalization — so applications don't need to build and maintain notification plumbing themselves.
Open Source, TypeScript, REST API, Redis, PostgreSQL, Notification Routing, Workflows
Gau Sampurna
Livestock Fertility & Cattle Health AI
Sperm Motility Analysis
Architected a containerized microservice for automated semen analysis. Engineered a high-performance ML pipeline integrating YOLOv8 for detection and DeepSORT for kinetic tracking via asynchronous Celery workers.
FastAPI, Celery, YOLOv8, DeepSORT, Redis, AWS S3
Gau Swastha — App Backend
Cattle health and advisory platform built by a team. My part: the structured-API backend (Gau-Gopalan-Seva) with cached GenAI answers and multilingual knowledge-base translations for the mobile app, plus multilingual weather advisories in the AI orchestration layer.
FastAPI, Google Gemini, PostgreSQL, AWS S3, Docker
Sumedha Design Systems
AI-Automation Internship · VLSI Documentation Automation
Other Projects
Full-stack, ML & research builds
NLP Health Tracker
Cloud-based dashboard integrating NLP for activity logging, processing 500+ complex food queries while reducing data entry time by 40%.
React.js, Node.js, NLP APIs, Google Sheets
Productivity Management
A full-stack MERN task-tracking system for manager–employee workflows, achieving real-time data synchronization with <50ms latency while supporting over 50+ concurrent active users.
Node.js, MongoDB Atlas, React, Google TTS, Render
Blogs
All postsHonest engineering notes: what I measured, what I got wrong, and what I would do differently.
A 5-line baseline humbled my 99.6%
I fine-tuned a 7B model to 270/271 on a held-out set. Then I trained TF-IDF + logistic regression on the same data and got 269/271. Here is what I changed in how I report results.
Tracking microscopic swimmers with YOLOv8 + DeepSORT on 2 vCPUs
How a microscopy video becomes a motility report: detection, tracking, a queue, and where the time actually goes.
How to backtest a price forecaster without fooling yourself
Your service's own error metric is probably measured on training data. Here is a rolling-origin backtest recipe, with a naive baseline in the room.
Cache your LLM like a database: bucket the inputs, template the output
When thousands of users ask almost the same question, don't pay for thousands of generations. A two-level template cache pattern.
When your validation set is secretly your training set
Fine-tuning on templated data makes leakage easy to create and easy to miss. The checks I run now, and how I report results when the leakage cannot be removed.
How I test an AI agent that must never invent a source
Sentinel drafts equity research with page-level citations, then audits itself for SEBI rules. Here is how I tested it, what I measured, and the number I stopped quoting.
Publications
Performance Evaluation of Deep Learning Models for Predicting Alzheimer's Disease
Research focused on the comparative analysis of DL architectures for early detection and prediction using neuroimaging data.
Enhancing Predictive Maintenance with SHAP and LIME: A Framework for Explainable AI
Developed a framework to make complex predictive maintenance models interpretable using XAI techniques for industrial applications.